initial commit, 4.5 stable

servers/rendering/renderer_rd/shaders/environment/SCsub

@@ -0,0 +1,18 @@
+#!/usr/bin/env python
+from misc.utility.scons_hints import *
+
+Import("env")
+
+if "RD_GLSL" in env["BUILDERS"]:
+    # find all include files
+    gl_include_files = [str(f) for f in Glob("*_inc.glsl")] + [str(f) for f in Glob("../*_inc.glsl")]
+
+    # find all shader code(all glsl files excluding our include files)
+    glsl_files = [str(f) for f in Glob("*.glsl") if str(f) not in gl_include_files]
+
+    # make sure we recompile shaders if include files change
+    env.Depends([f + ".gen.h" for f in glsl_files], gl_include_files + ["#glsl_builders.py"])
+
+    # compile shaders
+    for glsl_file in glsl_files:
+        env.RD_GLSL(glsl_file)

servers/rendering/renderer_rd/shaders/environment/gi.glsl

@@ -0,0 +1,788 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#ifdef SAMPLE_VOXEL_GI_NEAREST
+#extension GL_EXT_samplerless_texture_functions : enable
+#endif
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+#define M_PI 3.141592
+
+/* Specialization Constants (Toggles) */
+
+layout(constant_id = 0) const bool sc_half_res = false;
+layout(constant_id = 1) const bool sc_use_full_projection_matrix = false;
+layout(constant_id = 2) const bool sc_use_vrs = false;
+
+#define SDFGI_MAX_CASCADES 8
+
+//set 0 for SDFGI and render buffers
+
+layout(set = 0, binding = 1) uniform texture3D sdf_cascades[SDFGI_MAX_CASCADES];
+layout(set = 0, binding = 2) uniform texture3D light_cascades[SDFGI_MAX_CASCADES];
+layout(set = 0, binding = 3) uniform texture3D aniso0_cascades[SDFGI_MAX_CASCADES];
+layout(set = 0, binding = 4) uniform texture3D aniso1_cascades[SDFGI_MAX_CASCADES];
+layout(set = 0, binding = 5) uniform texture3D occlusion_texture;
+
+layout(set = 0, binding = 6) uniform sampler linear_sampler;
+layout(set = 0, binding = 7) uniform sampler linear_sampler_with_mipmaps;
+
+struct ProbeCascadeData {
+	vec3 position;
+	float to_probe;
+	ivec3 probe_world_offset;
+	float to_cell; // 1/bounds * grid_size
+	vec3 pad;
+	float exposure_normalization;
+};
+
+layout(rgba16f, set = 0, binding = 9) uniform restrict writeonly image2D ambient_buffer;
+layout(rgba16f, set = 0, binding = 10) uniform restrict writeonly image2D reflection_buffer;
+
+layout(set = 0, binding = 11) uniform texture2DArray lightprobe_texture;
+
+layout(set = 0, binding = 12) uniform texture2D depth_buffer;
+layout(set = 0, binding = 13) uniform texture2D normal_roughness_buffer;
+layout(set = 0, binding = 14) uniform utexture2D voxel_gi_buffer;
+
+layout(set = 0, binding = 15, std140) uniform SDFGI {
+	vec3 grid_size;
+	uint max_cascades;
+
+	bool use_occlusion;
+	int probe_axis_size;
+	float probe_to_uvw;
+	float normal_bias;
+
+	vec3 lightprobe_tex_pixel_size;
+	float energy;
+
+	vec3 lightprobe_uv_offset;
+	float y_mult;
+
+	vec3 occlusion_clamp;
+	uint pad3;
+
+	vec3 occlusion_renormalize;
+	uint pad4;
+
+	vec3 cascade_probe_size;
+	uint pad5;
+
+	ProbeCascadeData cascades[SDFGI_MAX_CASCADES];
+}
+sdfgi;
+
+#define MAX_VOXEL_GI_INSTANCES 8
+
+struct VoxelGIData {
+	mat4 xform; // 64 - 64
+
+	vec3 bounds; // 12 - 76
+	float dynamic_range; // 4 - 80
+
+	float bias; // 4 - 84
+	float normal_bias; // 4 - 88
+	bool blend_ambient; // 4 - 92
+	uint mipmaps; // 4 - 96
+
+	vec3 pad; // 12 - 108
+	float exposure_normalization; // 4 - 112
+};
+
+layout(set = 0, binding = 16, std140) uniform VoxelGIs {
+	VoxelGIData data[MAX_VOXEL_GI_INSTANCES];
+}
+voxel_gi_instances;
+
+layout(set = 0, binding = 17) uniform texture3D voxel_gi_textures[MAX_VOXEL_GI_INSTANCES];
+
+layout(set = 0, binding = 18, std140) uniform SceneData {
+	mat4x4 inv_projection[2];
+	mat4x4 cam_transform;
+	vec4 eye_offset[2];
+
+	ivec2 screen_size;
+	float pad1;
+	float pad2;
+}
+scene_data;
+
+#ifdef USE_VRS
+layout(r8ui, set = 0, binding = 19) uniform restrict readonly uimage2D vrs_buffer;
+#endif
+
+layout(push_constant, std430) uniform Params {
+	uint max_voxel_gi_instances;
+	bool high_quality_vct;
+	bool orthogonal;
+	uint view_index;
+
+	vec4 proj_info;
+
+	float z_near;
+	float z_far;
+	float pad2;
+	float pad3;
+}
+params;
+
+vec2 octahedron_wrap(vec2 v) {
+	vec2 signVal;
+	signVal.x = v.x >= 0.0 ? 1.0 : -1.0;
+	signVal.y = v.y >= 0.0 ? 1.0 : -1.0;
+	return (1.0 - abs(v.yx)) * signVal;
+}
+
+vec2 octahedron_encode(vec3 n) {
+	// https://twitter.com/Stubbesaurus/status/937994790553227264
+	n /= (abs(n.x) + abs(n.y) + abs(n.z));
+	n.xy = n.z >= 0.0 ? n.xy : octahedron_wrap(n.xy);
+	n.xy = n.xy * 0.5 + 0.5;
+	return n.xy;
+}
+
+vec4 blend_color(vec4 src, vec4 dst) {
+	vec4 res;
+	float sa = 1.0 - src.a;
+	res.a = dst.a * sa + src.a;
+	if (res.a == 0.0) {
+		res.rgb = vec3(0);
+	} else {
+		res.rgb = (dst.rgb * dst.a * sa + src.rgb * src.a) / res.a;
+	}
+	return res;
+}
+
+vec3 reconstruct_position(ivec2 screen_pos) {
+	if (sc_use_full_projection_matrix) {
+		vec4 pos;
+		pos.xy = (2.0 * vec2(screen_pos) / vec2(scene_data.screen_size)) - 1.0;
+		pos.z = texelFetch(sampler2D(depth_buffer, linear_sampler), screen_pos, 0).r * 2.0 - 1.0;
+		pos.w = 1.0;
+
+		pos = scene_data.inv_projection[params.view_index] * pos;
+
+		return pos.xyz / pos.w;
+	} else {
+		vec3 pos;
+		pos.z = texelFetch(sampler2D(depth_buffer, linear_sampler), screen_pos, 0).r;
+
+		pos.z = pos.z * 2.0 - 1.0;
+		if (params.orthogonal) {
+			pos.z = -(pos.z * (params.z_far - params.z_near) - (params.z_far + params.z_near)) / 2.0;
+		} else {
+			pos.z = 2.0 * params.z_near * params.z_far / (params.z_far + params.z_near + pos.z * (params.z_far - params.z_near));
+		}
+		pos.z = -pos.z;
+
+		pos.xy = vec2(screen_pos) * params.proj_info.xy + params.proj_info.zw;
+		if (!params.orthogonal) {
+			pos.xy *= pos.z;
+		}
+
+		return pos;
+	}
+}
+
+void sdfvoxel_gi_process(uint cascade, vec3 cascade_pos, vec3 cam_pos, vec3 cam_normal, vec3 cam_specular_normal, float roughness, out vec3 diffuse_light, out vec3 specular_light) {
+	cascade_pos += cam_normal * sdfgi.normal_bias;
+
+	vec3 base_pos = floor(cascade_pos);
+	//cascade_pos += mix(vec3(0.0),vec3(0.01),lessThan(abs(cascade_pos-base_pos),vec3(0.01))) * cam_normal;
+	ivec3 probe_base_pos = ivec3(base_pos);
+
+	vec4 diffuse_accum = vec4(0.0);
+	vec3 specular_accum;
+
+	ivec3 tex_pos = ivec3(probe_base_pos.xy, int(cascade));
+	tex_pos.x += probe_base_pos.z * sdfgi.probe_axis_size;
+	tex_pos.xy = tex_pos.xy * (SDFGI_OCT_SIZE + 2) + ivec2(1);
+
+	vec3 diffuse_posf = (vec3(tex_pos) + vec3(octahedron_encode(cam_normal) * float(SDFGI_OCT_SIZE), 0.0)) * sdfgi.lightprobe_tex_pixel_size;
+
+	vec3 specular_posf = (vec3(tex_pos) + vec3(octahedron_encode(cam_specular_normal) * float(SDFGI_OCT_SIZE), 0.0)) * sdfgi.lightprobe_tex_pixel_size;
+
+	specular_accum = vec3(0.0);
+
+	vec4 light_accum = vec4(0.0);
+	float weight_accum = 0.0;
+
+	for (uint j = 0; j < 8; j++) {
+		ivec3 offset = (ivec3(j) >> ivec3(0, 1, 2)) & ivec3(1, 1, 1);
+		ivec3 probe_posi = probe_base_pos;
+		probe_posi += offset;
+
+		// Compute weight
+
+		vec3 probe_pos = vec3(probe_posi);
+		vec3 probe_to_pos = cascade_pos - probe_pos;
+		vec3 probe_dir = normalize(-probe_to_pos);
+
+		vec3 trilinear = vec3(1.0) - abs(probe_to_pos);
+		float weight = trilinear.x * trilinear.y * trilinear.z * max(0.005, dot(cam_normal, probe_dir));
+
+		// Compute lightprobe occlusion
+
+		if (sdfgi.use_occlusion) {
+			ivec3 occ_indexv = abs((sdfgi.cascades[cascade].probe_world_offset + probe_posi) & ivec3(1, 1, 1)) * ivec3(1, 2, 4);
+			vec4 occ_mask = mix(vec4(0.0), vec4(1.0), equal(ivec4(occ_indexv.x | occ_indexv.y), ivec4(0, 1, 2, 3)));
+
+			vec3 occ_pos = clamp(cascade_pos, probe_pos - sdfgi.occlusion_clamp, probe_pos + sdfgi.occlusion_clamp) * sdfgi.probe_to_uvw;
+			occ_pos.z += float(cascade);
+			if (occ_indexv.z != 0) { //z bit is on, means index is >=4, so make it switch to the other half of textures
+				occ_pos.x += 1.0;
+			}
+
+			occ_pos *= sdfgi.occlusion_renormalize;
+			float occlusion = dot(textureLod(sampler3D(occlusion_texture, linear_sampler), occ_pos, 0.0), occ_mask);
+
+			weight *= max(occlusion, 0.01);
+		}
+
+		// Compute lightprobe texture position
+
+		vec3 diffuse;
+		vec3 pos_uvw = diffuse_posf;
+		pos_uvw.xy += vec2(offset.xy) * sdfgi.lightprobe_uv_offset.xy;
+		pos_uvw.x += float(offset.z) * sdfgi.lightprobe_uv_offset.z;
+		diffuse = textureLod(sampler2DArray(lightprobe_texture, linear_sampler), pos_uvw, 0.0).rgb;
+
+		diffuse_accum += vec4(diffuse * weight * sdfgi.cascades[cascade].exposure_normalization, weight);
+
+		{
+			vec3 specular = vec3(0.0);
+			vec3 pos_uvw = specular_posf;
+			pos_uvw.xy += vec2(offset.xy) * sdfgi.lightprobe_uv_offset.xy;
+			pos_uvw.x += float(offset.z) * sdfgi.lightprobe_uv_offset.z;
+			if (roughness < 0.99) {
+				specular = textureLod(sampler2DArray(lightprobe_texture, linear_sampler), pos_uvw + vec3(0, 0, float(sdfgi.max_cascades)), 0.0).rgb;
+			}
+			if (roughness > 0.2) {
+				specular = mix(specular, textureLod(sampler2DArray(lightprobe_texture, linear_sampler), pos_uvw, 0.0).rgb, (roughness - 0.2) * 1.25);
+			}
+
+			specular_accum += specular * weight * sdfgi.cascades[cascade].exposure_normalization;
+		}
+	}
+
+	if (diffuse_accum.a > 0.0) {
+		diffuse_accum.rgb /= diffuse_accum.a;
+	}
+
+	diffuse_light = diffuse_accum.rgb;
+
+	if (diffuse_accum.a > 0.0) {
+		specular_accum /= diffuse_accum.a;
+	}
+
+	specular_light = specular_accum;
+}
+
+void sdfgi_process(vec3 vertex, vec3 normal, vec3 reflection, float roughness, out vec4 ambient_light, out vec4 reflection_light) {
+	//make vertex orientation the world one, but still align to camera
+	vertex.y *= sdfgi.y_mult;
+	normal.y *= sdfgi.y_mult;
+	reflection.y *= sdfgi.y_mult;
+
+	//renormalize
+	normal = normalize(normal);
+	reflection = normalize(reflection);
+
+	vec3 cam_pos = vertex;
+	vec3 cam_normal = normal;
+
+	vec4 light_accum = vec4(0.0);
+	float weight_accum = 0.0;
+
+	vec4 light_blend_accum = vec4(0.0);
+	float weight_blend_accum = 0.0;
+
+	float blend = -1.0;
+
+	// helper constants, compute once
+
+	uint cascade = 0xFFFFFFFF;
+	vec3 cascade_pos;
+	vec3 cascade_normal;
+
+	for (uint i = 0; i < sdfgi.max_cascades; i++) {
+		cascade_pos = (cam_pos - sdfgi.cascades[i].position) * sdfgi.cascades[i].to_probe;
+
+		if (any(lessThan(cascade_pos, vec3(0.0))) || any(greaterThanEqual(cascade_pos, sdfgi.cascade_probe_size))) {
+			continue; //skip cascade
+		}
+
+		cascade = i;
+		break;
+	}
+
+	if (cascade < SDFGI_MAX_CASCADES) {
+		ambient_light = vec4(0, 0, 0, 1);
+		reflection_light = vec4(0, 0, 0, 1);
+
+		float blend;
+		vec3 diffuse, specular;
+		sdfvoxel_gi_process(cascade, cascade_pos, cam_pos, cam_normal, reflection, roughness, diffuse, specular);
+
+		{
+			//process blend
+			float blend_from = (float(sdfgi.probe_axis_size - 1) / 2.0) - 2.5;
+			float blend_to = blend_from + 2.0;
+
+			vec3 inner_pos = cam_pos * sdfgi.cascades[cascade].to_probe;
+
+			float len = length(inner_pos);
+
+			inner_pos = abs(normalize(inner_pos));
+			len *= max(inner_pos.x, max(inner_pos.y, inner_pos.z));
+
+			if (len >= blend_from) {
+				blend = smoothstep(blend_from, blend_to, len);
+			} else {
+				blend = 0.0;
+			}
+		}
+
+		if (blend > 0.0) {
+			//blend
+			if (cascade == sdfgi.max_cascades - 1) {
+				ambient_light.a = 1.0 - blend;
+				reflection_light.a = 1.0 - blend;
+
+			} else {
+				vec3 diffuse2, specular2;
+				cascade_pos = (cam_pos - sdfgi.cascades[cascade + 1].position) * sdfgi.cascades[cascade + 1].to_probe;
+				sdfvoxel_gi_process(cascade + 1, cascade_pos, cam_pos, cam_normal, reflection, roughness, diffuse2, specular2);
+				diffuse = mix(diffuse, diffuse2, blend);
+				specular = mix(specular, specular2, blend);
+			}
+		}
+
+		ambient_light.rgb = diffuse;
+
+		if (roughness < 0.2) {
+			vec3 pos_to_uvw = 1.0 / sdfgi.grid_size;
+			vec4 light_accum = vec4(0.0);
+
+			float blend_size = (sdfgi.grid_size.x / float(sdfgi.probe_axis_size - 1)) * 0.5;
+
+			float radius_sizes[SDFGI_MAX_CASCADES];
+			cascade = 0xFFFF;
+
+			float base_distance = length(cam_pos);
+			for (uint i = 0; i < sdfgi.max_cascades; i++) {
+				radius_sizes[i] = (1.0 / sdfgi.cascades[i].to_cell) * (sdfgi.grid_size.x * 0.5 - blend_size);
+				if (cascade == 0xFFFF && base_distance < radius_sizes[i]) {
+					cascade = i;
+				}
+			}
+
+			cascade = min(cascade, sdfgi.max_cascades - 1);
+
+			float max_distance = radius_sizes[sdfgi.max_cascades - 1];
+			vec3 ray_pos = cam_pos;
+			vec3 ray_dir = reflection;
+
+			{
+				float prev_radius = cascade > 0 ? radius_sizes[cascade - 1] : 0.0;
+				float base_blend = (base_distance - prev_radius) / (radius_sizes[cascade] - prev_radius);
+				float bias = (1.0 + base_blend) * 1.1;
+				vec3 abs_ray_dir = abs(ray_dir);
+				//ray_pos += ray_dir * (bias / sdfgi.cascades[cascade].to_cell); //bias to avoid self occlusion
+				ray_pos += (ray_dir * 1.0 / max(abs_ray_dir.x, max(abs_ray_dir.y, abs_ray_dir.z)) + cam_normal * 1.4) * bias / sdfgi.cascades[cascade].to_cell;
+			}
+			float softness = 0.2 + min(1.0, roughness * 5.0) * 4.0; //approximation to roughness so it does not seem like a hard fade
+			uint i = 0;
+			bool found = false;
+			while (true) {
+				if (length(ray_pos) >= max_distance || light_accum.a > 0.99) {
+					break;
+				}
+				if (!found && i >= cascade && length(ray_pos) < radius_sizes[i]) {
+					uint next_i = min(i + 1, sdfgi.max_cascades - 1);
+					cascade = max(i, cascade); //never go down
+
+					vec3 pos = ray_pos - sdfgi.cascades[i].position;
+					pos *= sdfgi.cascades[i].to_cell * pos_to_uvw;
+
+					float fdistance = textureLod(sampler3D(sdf_cascades[i], linear_sampler), pos, 0.0).r * 255.0 - 1.1;
+
+					vec4 hit_light = vec4(0.0);
+					if (fdistance < softness) {
+						hit_light.rgb = textureLod(sampler3D(light_cascades[i], linear_sampler), pos, 0.0).rgb;
+						hit_light.rgb *= 0.5; //approximation given value read is actually meant for anisotropy
+						hit_light.a = clamp(1.0 - (fdistance / softness), 0.0, 1.0);
+						hit_light.rgb *= hit_light.a;
+					}
+
+					fdistance /= sdfgi.cascades[i].to_cell;
+
+					if (i < (sdfgi.max_cascades - 1)) {
+						pos = ray_pos - sdfgi.cascades[next_i].position;
+						pos *= sdfgi.cascades[next_i].to_cell * pos_to_uvw;
+
+						float fdistance2 = textureLod(sampler3D(sdf_cascades[next_i], linear_sampler), pos, 0.0).r * 255.0 - 1.1;
+
+						vec4 hit_light2 = vec4(0.0);
+						if (fdistance2 < softness) {
+							hit_light2.rgb = textureLod(sampler3D(light_cascades[next_i], linear_sampler), pos, 0.0).rgb;
+							hit_light2.rgb *= 0.5; //approximation given value read is actually meant for anisotropy
+							hit_light2.a = clamp(1.0 - (fdistance2 / softness), 0.0, 1.0);
+							hit_light2.rgb *= hit_light2.a;
+						}
+
+						float prev_radius = i == 0 ? 0.0 : radius_sizes[max(0, i - 1)];
+						float blend = clamp((length(ray_pos) - prev_radius) / (radius_sizes[i] - prev_radius), 0.0, 1.0);
+
+						fdistance2 /= sdfgi.cascades[next_i].to_cell;
+
+						hit_light = mix(hit_light, hit_light2, blend);
+						fdistance = mix(fdistance, fdistance2, blend);
+					}
+
+					light_accum += hit_light;
+					ray_pos += ray_dir * fdistance;
+					found = true;
+				}
+				i++;
+				if (i == sdfgi.max_cascades) {
+					i = 0;
+					found = false;
+				}
+			}
+
+			vec3 light = light_accum.rgb / max(light_accum.a, 0.00001);
+			float alpha = min(1.0, light_accum.a);
+
+			float b = min(1.0, roughness * 5.0);
+
+			float sa = 1.0 - b;
+
+			reflection_light.a = alpha * sa + b;
+			if (reflection_light.a == 0) {
+				specular = vec3(0.0);
+			} else {
+				specular = (light * alpha * sa + specular * b) / reflection_light.a;
+			}
+		}
+
+		reflection_light.rgb = specular;
+
+		ambient_light.rgb *= sdfgi.energy;
+		reflection_light.rgb *= sdfgi.energy;
+	} else {
+		ambient_light = vec4(0);
+		reflection_light = vec4(0);
+	}
+}
+
+//standard voxel cone trace
+vec4 voxel_cone_trace(texture3D probe, vec3 cell_size, vec3 pos, vec3 direction, float tan_half_angle, float max_distance, float p_bias) {
+	float dist = p_bias;
+	vec4 color = vec4(0.0);
+
+	while (dist < max_distance && color.a < 0.95) {
+		float diameter = max(1.0, 2.0 * tan_half_angle * dist);
+		vec3 uvw_pos = (pos + dist * direction) * cell_size;
+		float half_diameter = diameter * 0.5;
+		//check if outside, then break
+		if (any(greaterThan(abs(uvw_pos - 0.5), vec3(0.5f + half_diameter * cell_size)))) {
+			break;
+		}
+		vec4 scolor = textureLod(sampler3D(probe, linear_sampler_with_mipmaps), uvw_pos, log2(diameter));
+		float a = (1.0 - color.a);
+		color += a * scolor;
+		dist += half_diameter;
+	}
+
+	return color;
+}
+
+vec4 voxel_cone_trace_45_degrees(texture3D probe, vec3 cell_size, vec3 pos, vec3 direction, float max_distance, float p_bias) {
+	float dist = p_bias;
+	vec4 color = vec4(0.0);
+	float radius = max(0.5, dist);
+	float lod_level = log2(radius * 2.0);
+
+	while (dist < max_distance && color.a < 0.95) {
+		vec3 uvw_pos = (pos + dist * direction) * cell_size;
+
+		//check if outside, then break
+		if (any(greaterThan(abs(uvw_pos - 0.5), vec3(0.5f + radius * cell_size)))) {
+			break;
+		}
+		vec4 scolor = textureLod(sampler3D(probe, linear_sampler_with_mipmaps), uvw_pos, lod_level);
+		lod_level += 1.0;
+
+		float a = (1.0 - color.a);
+		scolor *= a;
+		color += scolor;
+		dist += radius;
+		radius = max(0.5, dist);
+	}
+	return color;
+}
+
+void voxel_gi_compute(uint index, vec3 position, vec3 normal, vec3 ref_vec, mat3 normal_xform, float roughness, inout vec4 out_spec, inout vec4 out_diff, inout float out_blend) {
+	position = (voxel_gi_instances.data[index].xform * vec4(position, 1.0)).xyz;
+	ref_vec = normalize((voxel_gi_instances.data[index].xform * vec4(ref_vec, 0.0)).xyz);
+	normal = normalize((voxel_gi_instances.data[index].xform * vec4(normal, 0.0)).xyz);
+
+	position += normal * voxel_gi_instances.data[index].normal_bias;
+
+	//this causes corrupted pixels, i have no idea why..
+	if (any(bvec2(any(lessThan(position, vec3(0.0))), any(greaterThan(position, voxel_gi_instances.data[index].bounds))))) {
+		return;
+	}
+
+	mat3 dir_xform = mat3(voxel_gi_instances.data[index].xform) * normal_xform;
+
+	vec3 blendv = abs(position / voxel_gi_instances.data[index].bounds * 2.0 - 1.0);
+	float blend = clamp(1.0 - max(blendv.x, max(blendv.y, blendv.z)), 0.0, 1.0);
+	//float blend=1.0;
+
+	float max_distance = length(voxel_gi_instances.data[index].bounds);
+	vec3 cell_size = 1.0 / voxel_gi_instances.data[index].bounds;
+
+	//irradiance
+
+	vec4 light = vec4(0.0);
+
+	if (params.high_quality_vct) {
+		const uint cone_dir_count = 6;
+		vec3 cone_dirs[cone_dir_count] = vec3[](
+				vec3(0.0, 0.0, 1.0),
+				vec3(0.866025, 0.0, 0.5),
+				vec3(0.267617, 0.823639, 0.5),
+				vec3(-0.700629, 0.509037, 0.5),
+				vec3(-0.700629, -0.509037, 0.5),
+				vec3(0.267617, -0.823639, 0.5));
+
+		float cone_weights[cone_dir_count] = float[](0.25, 0.15, 0.15, 0.15, 0.15, 0.15);
+		float cone_angle_tan = 0.577;
+
+		for (uint i = 0; i < cone_dir_count; i++) {
+			vec3 dir = normalize(dir_xform * cone_dirs[i]);
+			light += cone_weights[i] * voxel_cone_trace(voxel_gi_textures[index], cell_size, position, dir, cone_angle_tan, max_distance, voxel_gi_instances.data[index].bias);
+		}
+	} else {
+		const uint cone_dir_count = 4;
+		vec3 cone_dirs[cone_dir_count] = vec3[](
+				vec3(0.707107, 0.0, 0.707107),
+				vec3(0.0, 0.707107, 0.707107),
+				vec3(-0.707107, 0.0, 0.707107),
+				vec3(0.0, -0.707107, 0.707107));
+
+		float cone_weights[cone_dir_count] = float[](0.25, 0.25, 0.25, 0.25);
+		for (int i = 0; i < cone_dir_count; i++) {
+			vec3 dir = normalize(dir_xform * cone_dirs[i]);
+			light += cone_weights[i] * voxel_cone_trace_45_degrees(voxel_gi_textures[index], cell_size, position, dir, max_distance, voxel_gi_instances.data[index].bias);
+		}
+	}
+
+	light.rgb *= voxel_gi_instances.data[index].dynamic_range * voxel_gi_instances.data[index].exposure_normalization;
+	if (!voxel_gi_instances.data[index].blend_ambient) {
+		light.a = 1.0;
+	}
+
+	out_diff += light * blend;
+
+	//radiance
+	vec4 irr_light = voxel_cone_trace(voxel_gi_textures[index], cell_size, position, ref_vec, tan(roughness * 0.5 * M_PI * 0.99), max_distance, voxel_gi_instances.data[index].bias);
+	irr_light.rgb *= voxel_gi_instances.data[index].dynamic_range * voxel_gi_instances.data[index].exposure_normalization;
+	if (!voxel_gi_instances.data[index].blend_ambient) {
+		irr_light.a = 1.0;
+	}
+
+	out_spec += irr_light * blend;
+
+	out_blend += blend;
+}
+
+vec4 fetch_normal_and_roughness(ivec2 pos) {
+	vec4 normal_roughness = texelFetch(sampler2D(normal_roughness_buffer, linear_sampler), pos, 0);
+	normal_roughness.xyz = normalize(normal_roughness.xyz * 2.0 - 1.0);
+	return normal_roughness;
+}
+
+void process_gi(ivec2 pos, vec3 vertex, inout vec4 ambient_light, inout vec4 reflection_light) {
+	vec4 normal_roughness = fetch_normal_and_roughness(pos);
+
+	vec3 normal = normal_roughness.xyz;
+
+	if (normal.length() > 0.5) {
+		//valid normal, can do GI
+		float roughness = normal_roughness.w;
+		bool dynamic_object = roughness > 0.5;
+		if (dynamic_object) {
+			roughness = 1.0 - roughness;
+		}
+		roughness /= (127.0 / 255.0);
+		vec3 view = -normalize(mat3(scene_data.cam_transform) * (vertex - scene_data.eye_offset[gl_GlobalInvocationID.z].xyz));
+		vertex = mat3(scene_data.cam_transform) * vertex;
+		normal = normalize(mat3(scene_data.cam_transform) * normal);
+		vec3 reflection = normalize(reflect(-view, normal));
+
+#ifdef USE_SDFGI
+		sdfgi_process(vertex, normal, reflection, roughness, ambient_light, reflection_light);
+#endif
+
+#ifdef USE_VOXEL_GI_INSTANCES
+		{
+#ifdef SAMPLE_VOXEL_GI_NEAREST
+			uvec2 voxel_gi_tex = texelFetch(voxel_gi_buffer, pos, 0).rg;
+#else
+			uvec2 voxel_gi_tex = texelFetch(usampler2D(voxel_gi_buffer, linear_sampler), pos, 0).rg;
+#endif
+			roughness *= roughness;
+			//find arbitrary tangent and bitangent, then build a matrix
+			vec3 v0 = abs(normal.z) < 0.999 ? vec3(0.0, 0.0, 1.0) : vec3(0.0, 1.0, 0.0);
+			vec3 tangent = normalize(cross(v0, normal));
+			vec3 bitangent = normalize(cross(tangent, normal));
+			mat3 normal_mat = mat3(tangent, bitangent, normal);
+
+			vec4 amb_accum = vec4(0.0);
+			vec4 spec_accum = vec4(0.0);
+			float blend_accum = 0.0;
+
+			for (uint i = 0; i < params.max_voxel_gi_instances; i++) {
+				if (any(equal(uvec2(i), voxel_gi_tex))) {
+					voxel_gi_compute(i, vertex, normal, reflection, normal_mat, roughness, spec_accum, amb_accum, blend_accum);
+				}
+			}
+			if (blend_accum > 0.0) {
+				amb_accum /= blend_accum;
+				spec_accum /= blend_accum;
+			}
+
+#ifdef USE_SDFGI
+			reflection_light = blend_color(spec_accum, reflection_light);
+			ambient_light = blend_color(amb_accum, ambient_light);
+#else
+			reflection_light = spec_accum;
+			ambient_light = amb_accum;
+#endif
+		}
+#endif
+	}
+}
+
+void main() {
+	ivec2 pos = ivec2(gl_GlobalInvocationID.xy);
+
+	uint vrs_x, vrs_y;
+#ifdef USE_VRS
+	if (sc_use_vrs) {
+		ivec2 vrs_pos;
+
+		// Currently we use a 16x16 texel, possibly some day make this configurable.
+		if (sc_half_res) {
+			vrs_pos = pos >> 3;
+		} else {
+			vrs_pos = pos >> 4;
+		}
+
+		uint vrs_texel = imageLoad(vrs_buffer, vrs_pos).r;
+		// note, valid values for vrs_x and vrs_y are 1, 2 and 4.
+		vrs_x = 1 << ((vrs_texel >> 2) & 3);
+		vrs_y = 1 << (vrs_texel & 3);
+
+		if (mod(pos.x, vrs_x) != 0) {
+			return;
+		}
+
+		if (mod(pos.y, vrs_y) != 0) {
+			return;
+		}
+	}
+#endif
+
+	if (sc_half_res) {
+		pos <<= 1;
+	}
+
+	if (any(greaterThanEqual(pos, scene_data.screen_size))) { //too large, do nothing
+		return;
+	}
+
+	vec4 ambient_light = vec4(0.0);
+	vec4 reflection_light = vec4(0.0);
+
+	vec3 vertex = reconstruct_position(pos);
+	vertex.y = -vertex.y;
+
+	process_gi(pos, vertex, ambient_light, reflection_light);
+
+	if (sc_half_res) {
+		pos >>= 1;
+	}
+
+	imageStore(ambient_buffer, pos, ambient_light);
+	imageStore(reflection_buffer, pos, reflection_light);
+
+#ifdef USE_VRS
+	if (sc_use_vrs) {
+		if (vrs_x > 1) {
+			imageStore(ambient_buffer, pos + ivec2(1, 0), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(1, 0), reflection_light);
+		}
+
+		if (vrs_x > 2) {
+			imageStore(ambient_buffer, pos + ivec2(2, 0), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(2, 0), reflection_light);
+
+			imageStore(ambient_buffer, pos + ivec2(3, 0), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(3, 0), reflection_light);
+		}
+
+		if (vrs_y > 1) {
+			imageStore(ambient_buffer, pos + ivec2(0, 1), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(0, 1), reflection_light);
+		}
+
+		if (vrs_y > 1 && vrs_x > 1) {
+			imageStore(ambient_buffer, pos + ivec2(1, 1), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(1, 1), reflection_light);
+		}
+
+		if (vrs_y > 1 && vrs_x > 2) {
+			imageStore(ambient_buffer, pos + ivec2(2, 1), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(2, 1), reflection_light);
+
+			imageStore(ambient_buffer, pos + ivec2(3, 1), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(3, 1), reflection_light);
+		}
+
+		if (vrs_y > 2) {
+			imageStore(ambient_buffer, pos + ivec2(0, 2), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(0, 2), reflection_light);
+			imageStore(ambient_buffer, pos + ivec2(0, 3), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(0, 3), reflection_light);
+		}
+
+		if (vrs_y > 2 && vrs_x > 1) {
+			imageStore(ambient_buffer, pos + ivec2(1, 2), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(1, 2), reflection_light);
+			imageStore(ambient_buffer, pos + ivec2(1, 3), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(1, 3), reflection_light);
+		}
+
+		if (vrs_y > 2 && vrs_x > 2) {
+			imageStore(ambient_buffer, pos + ivec2(2, 2), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(2, 2), reflection_light);
+			imageStore(ambient_buffer, pos + ivec2(2, 3), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(2, 3), reflection_light);
+
+			imageStore(ambient_buffer, pos + ivec2(3, 2), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(3, 2), reflection_light);
+			imageStore(ambient_buffer, pos + ivec2(3, 3), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(3, 3), reflection_light);
+		}
+	}
+#endif
+}

servers/rendering/renderer_rd/shaders/environment/sdfgi_debug.glsl

@@ -0,0 +1,185 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+#define MAX_CASCADES 8
+
+layout(set = 0, binding = 1) uniform texture3D sdf_cascades[MAX_CASCADES];
+layout(set = 0, binding = 2) uniform texture3D light_cascades[MAX_CASCADES];
+layout(set = 0, binding = 3) uniform texture3D aniso0_cascades[MAX_CASCADES];
+layout(set = 0, binding = 4) uniform texture3D aniso1_cascades[MAX_CASCADES];
+layout(set = 0, binding = 5) uniform texture3D occlusion_texture;
+
+layout(set = 0, binding = 8) uniform sampler linear_sampler;
+
+struct CascadeData {
+	vec3 offset; //offset of (0,0,0) in world coordinates
+	float to_cell; // 1/bounds * grid_size
+	ivec3 probe_world_offset;
+	uint pad;
+	vec4 pad2;
+};
+
+layout(set = 0, binding = 9, std140) uniform Cascades {
+	CascadeData data[MAX_CASCADES];
+}
+cascades;
+
+layout(rgba16f, set = 0, binding = 10) uniform restrict writeonly image2D screen_buffer;
+
+layout(set = 0, binding = 11) uniform texture2DArray lightprobe_texture;
+
+layout(push_constant, std430) uniform Params {
+	vec3 grid_size;
+	uint max_cascades;
+
+	ivec2 screen_size;
+	float y_mult;
+
+	float z_near;
+
+	mat3x4 inv_projection;
+	// We pack these more tightly than mat3 and vec3, which will require some reconstruction trickery.
+	float cam_basis[3][3];
+	float cam_origin[3];
+}
+params;
+
+vec3 linear_to_srgb(vec3 color) {
+	//if going to srgb, clamp from 0 to 1.
+	color = clamp(color, vec3(0.0), vec3(1.0));
+	const vec3 a = vec3(0.055f);
+	return mix((vec3(1.0f) + a) * pow(color.rgb, vec3(1.0f / 2.4f)) - a, 12.92f * color.rgb, lessThan(color.rgb, vec3(0.0031308f)));
+}
+
+vec2 octahedron_wrap(vec2 v) {
+	vec2 signVal;
+	signVal.x = v.x >= 0.0 ? 1.0 : -1.0;
+	signVal.y = v.y >= 0.0 ? 1.0 : -1.0;
+	return (1.0 - abs(v.yx)) * signVal;
+}
+
+vec2 octahedron_encode(vec3 n) {
+	// https://twitter.com/Stubbesaurus/status/937994790553227264
+	n /= (abs(n.x) + abs(n.y) + abs(n.z));
+	n.xy = n.z >= 0.0 ? n.xy : octahedron_wrap(n.xy);
+	n.xy = n.xy * 0.5 + 0.5;
+	return n.xy;
+}
+
+void main() {
+	// Pixel being shaded
+	ivec2 screen_pos = ivec2(gl_GlobalInvocationID.xy);
+	if (any(greaterThanEqual(screen_pos, params.screen_size))) { //too large, do nothing
+		return;
+	}
+
+	vec3 ray_pos;
+	vec3 ray_dir;
+	{
+		ray_pos = vec3(params.cam_origin[0], params.cam_origin[1], params.cam_origin[2]);
+
+		ray_dir.xy = ((vec2(screen_pos) / vec2(params.screen_size)) * 2.0 - 1.0);
+		ray_dir.z = params.z_near;
+
+		ray_dir = (vec4(ray_dir, 1.0) * mat4(params.inv_projection)).xyz;
+
+		mat3 cam_basis;
+		{
+			vec3 c0 = vec3(params.cam_basis[0][0], params.cam_basis[0][1], params.cam_basis[0][2]);
+			vec3 c1 = vec3(params.cam_basis[1][0], params.cam_basis[1][1], params.cam_basis[1][2]);
+			vec3 c2 = vec3(params.cam_basis[2][0], params.cam_basis[2][1], params.cam_basis[2][2]);
+			cam_basis = mat3(c0, c1, c2);
+		}
+		ray_dir = normalize(cam_basis * ray_dir);
+	}
+
+	ray_pos.y *= params.y_mult;
+	ray_dir.y *= params.y_mult;
+	ray_dir = normalize(ray_dir);
+
+	vec3 pos_to_uvw = 1.0 / params.grid_size;
+
+	vec3 light = vec3(0.0);
+	float blend = 0.0;
+
+#if 1
+	// No interpolation
+
+	vec3 inv_dir = 1.0 / ray_dir;
+
+	float rough = 0.5;
+	bool hit = false;
+
+	for (uint i = 0; i < params.max_cascades; i++) {
+		//convert to local bounds
+		vec3 pos = ray_pos - cascades.data[i].offset;
+		pos *= cascades.data[i].to_cell;
+
+		// Should never happen for debug, since we start mostly at the bounds center,
+		// but add anyway.
+		//if (any(lessThan(pos,vec3(0.0))) || any(greaterThanEqual(pos,params.grid_size))) {
+		//	continue; //already past bounds for this cascade, goto next
+		//}
+
+		//find maximum advance distance (until reaching bounds)
+		vec3 t0 = -pos * inv_dir;
+		vec3 t1 = (params.grid_size - pos) * inv_dir;
+		vec3 tmax = max(t0, t1);
+		float max_advance = min(tmax.x, min(tmax.y, tmax.z));
+
+		float advance = 0.0;
+		vec3 uvw;
+		hit = false;
+
+		while (advance < max_advance) {
+			//read how much to advance from SDF
+			uvw = (pos + ray_dir * advance) * pos_to_uvw;
+
+			float distance = texture(sampler3D(sdf_cascades[i], linear_sampler), uvw).r * 255.0 - 1.7;
+
+			if (distance < 0.001) {
+				//consider hit
+				hit = true;
+				break;
+			}
+
+			advance += distance;
+		}
+
+		if (!hit) {
+			pos += ray_dir * min(advance, max_advance);
+			pos /= cascades.data[i].to_cell;
+			pos += cascades.data[i].offset;
+			ray_pos = pos;
+			continue;
+		}
+
+		//compute albedo, emission and normal at hit point
+
+		const float EPSILON = 0.001;
+		vec3 hit_normal = normalize(vec3(
+				texture(sampler3D(sdf_cascades[i], linear_sampler), uvw + vec3(EPSILON, 0.0, 0.0)).r - texture(sampler3D(sdf_cascades[i], linear_sampler), uvw - vec3(EPSILON, 0.0, 0.0)).r,
+				texture(sampler3D(sdf_cascades[i], linear_sampler), uvw + vec3(0.0, EPSILON, 0.0)).r - texture(sampler3D(sdf_cascades[i], linear_sampler), uvw - vec3(0.0, EPSILON, 0.0)).r,
+				texture(sampler3D(sdf_cascades[i], linear_sampler), uvw + vec3(0.0, 0.0, EPSILON)).r - texture(sampler3D(sdf_cascades[i], linear_sampler), uvw - vec3(0.0, 0.0, EPSILON)).r));
+
+		vec3 hit_light = texture(sampler3D(light_cascades[i], linear_sampler), uvw).rgb;
+		vec4 aniso0 = texture(sampler3D(aniso0_cascades[i], linear_sampler), uvw);
+		vec3 hit_aniso0 = aniso0.rgb;
+		vec3 hit_aniso1 = vec3(aniso0.a, texture(sampler3D(aniso1_cascades[i], linear_sampler), uvw).rg);
+
+		hit_light *= (dot(max(vec3(0.0), (hit_normal * hit_aniso0)), vec3(1.0)) + dot(max(vec3(0.0), (-hit_normal * hit_aniso1)), vec3(1.0)));
+
+		light = hit_light;
+
+		break;
+	}
+
+#endif
+
+	imageStore(screen_buffer, screen_pos, vec4(linear_to_srgb(light), 1.0));
+}

servers/rendering/renderer_rd/shaders/environment/sdfgi_debug_probes.glsl

@@ -0,0 +1,243 @@
+#[vertex]
+
+#version 450
+
+#ifdef USE_MULTIVIEW
+#extension GL_EXT_multiview : enable
+#define ViewIndex gl_ViewIndex
+#else // USE_MULTIVIEW
+#define ViewIndex 0
+#endif // !USE_MULTIVIEW
+
+#VERSION_DEFINES
+
+#define MAX_CASCADES 8
+#define MAX_VIEWS 2
+
+layout(push_constant, std430) uniform Params {
+	uint band_power;
+	uint sections_in_band;
+	uint band_mask;
+	float section_arc;
+
+	vec3 grid_size;
+	uint cascade;
+
+	uint pad;
+	float y_mult;
+	uint probe_debug_index;
+	int probe_axis_size;
+}
+params;
+
+// https://in4k.untergrund.net/html_articles/hugi_27_-_coding_corner_polaris_sphere_tessellation_101.htm
+
+vec3 get_sphere_vertex(uint p_vertex_id) {
+	float x_angle = float(p_vertex_id & 1u) + (p_vertex_id >> params.band_power);
+
+	float y_angle =
+			float((p_vertex_id & params.band_mask) >> 1) + ((p_vertex_id >> params.band_power) * params.sections_in_band);
+
+	x_angle *= params.section_arc * 0.5f; // remember - 180AA x rot not 360
+	y_angle *= -params.section_arc;
+
+	vec3 point = vec3(sin(x_angle) * sin(y_angle), cos(x_angle), sin(x_angle) * cos(y_angle));
+
+	return point;
+}
+
+#ifdef MODE_PROBES
+
+layout(location = 0) out vec3 normal_interp;
+layout(location = 1) out flat uint probe_index;
+
+#endif
+
+#ifdef MODE_VISIBILITY
+
+layout(location = 0) out float visibility;
+
+#endif
+
+struct CascadeData {
+	vec3 offset; //offset of (0,0,0) in world coordinates
+	float to_cell; // 1/bounds * grid_size
+	ivec3 probe_world_offset;
+	uint pad;
+	vec4 pad2;
+};
+
+layout(set = 0, binding = 1, std140) uniform Cascades {
+	CascadeData data[MAX_CASCADES];
+}
+cascades;
+
+layout(set = 0, binding = 4) uniform texture3D occlusion_texture;
+layout(set = 0, binding = 3) uniform sampler linear_sampler;
+
+layout(set = 0, binding = 5, std140) uniform SceneData {
+	mat4 projection[MAX_VIEWS];
+}
+scene_data;
+
+void main() {
+#ifdef MODE_PROBES
+	probe_index = gl_InstanceIndex;
+
+	normal_interp = get_sphere_vertex(gl_VertexIndex);
+
+	vec3 vertex = normal_interp * 0.2;
+
+	float probe_cell_size = float(params.grid_size / float(params.probe_axis_size - 1)) / cascades.data[params.cascade].to_cell;
+
+	ivec3 probe_cell;
+	probe_cell.x = int(probe_index % params.probe_axis_size);
+	probe_cell.y = int(probe_index / (params.probe_axis_size * params.probe_axis_size));
+	probe_cell.z = int((probe_index / params.probe_axis_size) % params.probe_axis_size);
+
+	vertex += (cascades.data[params.cascade].offset + vec3(probe_cell) * probe_cell_size) / vec3(1.0, params.y_mult, 1.0);
+
+	gl_Position = scene_data.projection[ViewIndex] * vec4(vertex, 1.0);
+#endif
+
+#ifdef MODE_VISIBILITY
+
+	int probe_index = int(params.probe_debug_index);
+
+	vec3 vertex = get_sphere_vertex(gl_VertexIndex) * 0.01;
+
+	float probe_cell_size = float(params.grid_size / float(params.probe_axis_size - 1)) / cascades.data[params.cascade].to_cell;
+
+	ivec3 probe_cell;
+	probe_cell.x = int(probe_index % params.probe_axis_size);
+	probe_cell.y = int((probe_index % (params.probe_axis_size * params.probe_axis_size)) / params.probe_axis_size);
+	probe_cell.z = int(probe_index / (params.probe_axis_size * params.probe_axis_size));
+
+	vertex += (cascades.data[params.cascade].offset + vec3(probe_cell) * probe_cell_size) / vec3(1.0, params.y_mult, 1.0);
+
+	int probe_voxels = int(params.grid_size.x) / int(params.probe_axis_size - 1);
+	int occluder_index = int(gl_InstanceIndex);
+
+	int diameter = probe_voxels * 2;
+	ivec3 occluder_pos;
+	occluder_pos.x = int(occluder_index % diameter);
+	occluder_pos.y = int(occluder_index / (diameter * diameter));
+	occluder_pos.z = int((occluder_index / diameter) % diameter);
+
+	float cell_size = 1.0 / cascades.data[params.cascade].to_cell;
+
+	ivec3 occluder_offset = occluder_pos - ivec3(diameter / 2);
+	vertex += ((vec3(occluder_offset) + vec3(0.5)) * cell_size) / vec3(1.0, params.y_mult, 1.0);
+
+	ivec3 global_cell = probe_cell + cascades.data[params.cascade].probe_world_offset;
+	uint occlusion_layer = 0;
+	if ((global_cell.x & 1) != 0) {
+		occlusion_layer |= 1;
+	}
+	if ((global_cell.y & 1) != 0) {
+		occlusion_layer |= 2;
+	}
+	if ((global_cell.z & 1) != 0) {
+		occlusion_layer |= 4;
+	}
+	ivec3 tex_pos = probe_cell * probe_voxels + occluder_offset;
+
+	const vec4 layer_axis[4] = vec4[](
+			vec4(1, 0, 0, 0),
+			vec4(0, 1, 0, 0),
+			vec4(0, 0, 1, 0),
+			vec4(0, 0, 0, 1));
+
+	tex_pos.z += int(params.cascade) * int(params.grid_size);
+	if (occlusion_layer >= 4) {
+		tex_pos.x += int(params.grid_size.x);
+		occlusion_layer &= 3;
+	}
+
+	visibility = dot(texelFetch(sampler3D(occlusion_texture, linear_sampler), tex_pos, 0), layer_axis[occlusion_layer]);
+
+	gl_Position = scene_data.projection[ViewIndex] * vec4(vertex, 1.0);
+
+#endif
+}
+
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+#define MAX_VIEWS 2
+
+layout(location = 0) out vec4 frag_color;
+
+layout(set = 0, binding = 2) uniform texture2DArray lightprobe_texture;
+layout(set = 0, binding = 3) uniform sampler linear_sampler;
+
+layout(push_constant, std430) uniform Params {
+	uint band_power;
+	uint sections_in_band;
+	uint band_mask;
+	float section_arc;
+
+	vec3 grid_size;
+	uint cascade;
+
+	uint pad;
+	float y_mult;
+	uint probe_debug_index;
+	int probe_axis_size;
+}
+params;
+
+#ifdef MODE_PROBES
+
+layout(location = 0) in vec3 normal_interp;
+layout(location = 1) in flat uint probe_index;
+
+#endif
+
+#ifdef MODE_VISIBILITY
+layout(location = 0) in float visibility;
+#endif
+
+vec2 octahedron_wrap(vec2 v) {
+	vec2 signVal;
+	signVal.x = v.x >= 0.0 ? 1.0 : -1.0;
+	signVal.y = v.y >= 0.0 ? 1.0 : -1.0;
+	return (1.0 - abs(v.yx)) * signVal;
+}
+
+vec2 octahedron_encode(vec3 n) {
+	// https://twitter.com/Stubbesaurus/status/937994790553227264
+	n /= (abs(n.x) + abs(n.y) + abs(n.z));
+	n.xy = n.z >= 0.0 ? n.xy : octahedron_wrap(n.xy);
+	n.xy = n.xy * 0.5 + 0.5;
+	return n.xy;
+}
+
+void main() {
+#ifdef MODE_PROBES
+
+	ivec3 tex_pos;
+	tex_pos.x = int(probe_index) % params.probe_axis_size; //x
+	tex_pos.y = int(probe_index) / (params.probe_axis_size * params.probe_axis_size);
+	tex_pos.x += params.probe_axis_size * ((int(probe_index) / params.probe_axis_size) % params.probe_axis_size); //z
+	tex_pos.z = int(params.cascade);
+
+	vec3 tex_pos_ofs = vec3(octahedron_encode(normal_interp) * float(OCT_SIZE), 0.0);
+	vec3 tex_posf = vec3(vec2(tex_pos.xy * (OCT_SIZE + 2) + ivec2(1)), float(tex_pos.z)) + tex_pos_ofs;
+
+	tex_posf.xy /= vec2(ivec2(params.probe_axis_size * params.probe_axis_size * (OCT_SIZE + 2), params.probe_axis_size * (OCT_SIZE + 2)));
+
+	vec4 indirect_light = textureLod(sampler2DArray(lightprobe_texture, linear_sampler), tex_posf, 0.0);
+
+	frag_color = indirect_light;
+
+#endif
+
+#ifdef MODE_VISIBILITY
+
+	frag_color = vec4(vec3(1, visibility, visibility), 1.0);
+#endif
+}

servers/rendering/renderer_rd/shaders/environment/sdfgi_direct_light.glsl

@@ -0,0 +1,507 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
+
+#define MAX_CASCADES 8
+
+layout(set = 0, binding = 1) uniform texture3D sdf_cascades[MAX_CASCADES];
+layout(set = 0, binding = 2) uniform sampler linear_sampler;
+
+layout(set = 0, binding = 3, std430) restrict readonly buffer DispatchData {
+	uint x;
+	uint y;
+	uint z;
+	uint total_count;
+}
+dispatch_data;
+
+struct ProcessVoxel {
+	uint position; // xyz 7 bit packed, extra 11 bits for neighbors.
+	uint albedo; // rgb bits 0-15 albedo, bits 16-21 are normal bits (set if geometry exists toward that side), extra 11 bits for neighbors.
+	uint light; // rgbe8985 encoded total saved light, extra 2 bits for neighbors.
+	uint light_aniso; // 55555 light anisotropy, extra 2 bits for neighbors.
+	//total neighbors: 26
+};
+
+#ifdef MODE_PROCESS_STATIC
+layout(set = 0, binding = 4, std430) restrict buffer ProcessVoxels {
+#else
+layout(set = 0, binding = 4, std430) restrict buffer readonly ProcessVoxels {
+#endif
+	ProcessVoxel data[];
+}
+process_voxels;
+
+layout(r32ui, set = 0, binding = 5) uniform restrict uimage3D dst_light;
+layout(rgba8, set = 0, binding = 6) uniform restrict image3D dst_aniso0;
+layout(rg8, set = 0, binding = 7) uniform restrict image3D dst_aniso1;
+
+struct CascadeData {
+	vec3 offset; //offset of (0,0,0) in world coordinates
+	float to_cell; // 1/bounds * grid_size
+	ivec3 probe_world_offset;
+	uint pad;
+	vec4 pad2;
+};
+
+layout(set = 0, binding = 8, std140) uniform Cascades {
+	CascadeData data[MAX_CASCADES];
+}
+cascades;
+
+#define LIGHT_TYPE_DIRECTIONAL 0
+#define LIGHT_TYPE_OMNI 1
+#define LIGHT_TYPE_SPOT 2
+
+struct Light {
+	vec3 color;
+	float energy;
+
+	vec3 direction;
+	bool has_shadow;
+
+	vec3 position;
+	float attenuation;
+
+	uint type;
+	float cos_spot_angle;
+	float inv_spot_attenuation;
+	float radius;
+};
+
+layout(set = 0, binding = 9, std140) buffer restrict readonly Lights {
+	Light data[];
+}
+lights;
+
+layout(set = 0, binding = 10) uniform texture2DArray lightprobe_texture;
+layout(set = 0, binding = 11) uniform texture3D occlusion_texture;
+
+layout(push_constant, std430) uniform Params {
+	vec3 grid_size;
+	uint max_cascades;
+
+	uint cascade;
+	uint light_count;
+	uint process_offset;
+	uint process_increment;
+
+	int probe_axis_size;
+	float bounce_feedback;
+	float y_mult;
+	bool use_occlusion;
+}
+params;
+
+vec2 octahedron_wrap(vec2 v) {
+	vec2 signVal;
+	signVal.x = v.x >= 0.0 ? 1.0 : -1.0;
+	signVal.y = v.y >= 0.0 ? 1.0 : -1.0;
+	return (1.0 - abs(v.yx)) * signVal;
+}
+
+vec2 octahedron_encode(vec3 n) {
+	// https://twitter.com/Stubbesaurus/status/937994790553227264
+	n /= (abs(n.x) + abs(n.y) + abs(n.z));
+	n.xy = n.z >= 0.0 ? n.xy : octahedron_wrap(n.xy);
+	n.xy = n.xy * 0.5 + 0.5;
+	return n.xy;
+}
+
+float get_omni_attenuation(float distance, float inv_range, float decay) {
+	float nd = distance * inv_range;
+	nd *= nd;
+	nd *= nd; // nd^4
+	nd = max(1.0 - nd, 0.0);
+	nd *= nd; // nd^2
+	return nd * pow(max(distance, 0.0001), -decay);
+}
+
+void main() {
+	uint voxel_index = uint(gl_GlobalInvocationID.x);
+
+	//used for skipping voxels every N frames
+	if (params.process_increment > 1) {
+		voxel_index *= params.process_increment;
+		voxel_index += params.process_offset;
+	}
+
+	if (voxel_index >= dispatch_data.total_count) {
+		return;
+	}
+
+	uint voxel_position = process_voxels.data[voxel_index].position;
+
+	//keep for storing to texture
+	ivec3 positioni = ivec3((uvec3(voxel_position, voxel_position, voxel_position) >> uvec3(0, 7, 14)) & uvec3(0x7F));
+
+	vec3 position = vec3(positioni) + vec3(0.5);
+	position /= cascades.data[params.cascade].to_cell;
+	position += cascades.data[params.cascade].offset;
+
+	uint voxel_albedo = process_voxels.data[voxel_index].albedo;
+
+	vec3 albedo = vec3(uvec3(voxel_albedo >> 10, voxel_albedo >> 5, voxel_albedo) & uvec3(0x1F)) / float(0x1F);
+	vec3 light_accum[6] = vec3[](vec3(0.0), vec3(0.0), vec3(0.0), vec3(0.0), vec3(0.0), vec3(0.0));
+	uint valid_aniso = (voxel_albedo >> 15) & 0x3F;
+
+	const vec3 aniso_dir[6] = vec3[](
+			vec3(1, 0, 0),
+			vec3(0, 1, 0),
+			vec3(0, 0, 1),
+			vec3(-1, 0, 0),
+			vec3(0, -1, 0),
+			vec3(0, 0, -1));
+
+	// Add indirect light first, in order to save computation resources
+#ifdef MODE_PROCESS_DYNAMIC
+	if (params.bounce_feedback > 0.001) {
+		vec3 feedback = (params.bounce_feedback < 1.0) ? (albedo * params.bounce_feedback) : mix(albedo, vec3(1.0), params.bounce_feedback - 1.0);
+		vec3 pos = (vec3(positioni) + vec3(0.5)) * float(params.probe_axis_size - 1) / params.grid_size;
+		ivec3 probe_base_pos = ivec3(pos);
+
+		float weight_accum[6] = float[](0, 0, 0, 0, 0, 0);
+
+		ivec3 tex_pos = ivec3(probe_base_pos.xy, int(params.cascade));
+		tex_pos.x += probe_base_pos.z * int(params.probe_axis_size);
+
+		tex_pos.xy = tex_pos.xy * (OCT_SIZE + 2) + ivec2(1);
+
+		vec3 base_tex_posf = vec3(tex_pos);
+		vec2 tex_pixel_size = 1.0 / vec2(ivec2((OCT_SIZE + 2) * params.probe_axis_size * params.probe_axis_size, (OCT_SIZE + 2) * params.probe_axis_size));
+		vec3 probe_uv_offset = vec3(ivec3(OCT_SIZE + 2, OCT_SIZE + 2, (OCT_SIZE + 2) * params.probe_axis_size)) * tex_pixel_size.xyx;
+
+		for (uint j = 0; j < 8; j++) {
+			ivec3 offset = (ivec3(j) >> ivec3(0, 1, 2)) & ivec3(1, 1, 1);
+			ivec3 probe_posi = probe_base_pos;
+			probe_posi += offset;
+
+			// Compute weight
+
+			vec3 probe_pos = vec3(probe_posi);
+			vec3 probe_to_pos = pos - probe_pos;
+			vec3 probe_dir = normalize(-probe_to_pos);
+
+			// Compute lightprobe texture position
+
+			vec3 trilinear = vec3(1.0) - abs(probe_to_pos);
+
+			for (uint k = 0; k < 6; k++) {
+				if (bool(valid_aniso & (1 << k))) {
+					vec3 n = aniso_dir[k];
+					float weight = trilinear.x * trilinear.y * trilinear.z * max(0, dot(n, probe_dir));
+
+					if (weight > 0.0 && params.use_occlusion) {
+						ivec3 occ_indexv = abs((cascades.data[params.cascade].probe_world_offset + probe_posi) & ivec3(1, 1, 1)) * ivec3(1, 2, 4);
+						vec4 occ_mask = mix(vec4(0.0), vec4(1.0), equal(ivec4(occ_indexv.x | occ_indexv.y), ivec4(0, 1, 2, 3)));
+
+						vec3 occ_pos = (vec3(positioni) + aniso_dir[k] + vec3(0.5)) / params.grid_size;
+						occ_pos.z += float(params.cascade);
+						if (occ_indexv.z != 0) { //z bit is on, means index is >=4, so make it switch to the other half of textures
+							occ_pos.x += 1.0;
+						}
+						occ_pos *= vec3(0.5, 1.0, 1.0 / float(params.max_cascades)); //renormalize
+						float occlusion = dot(textureLod(sampler3D(occlusion_texture, linear_sampler), occ_pos, 0.0), occ_mask);
+
+						weight *= occlusion;
+					}
+
+					if (weight > 0.0) {
+						vec3 tex_posf = base_tex_posf + vec3(octahedron_encode(n) * float(OCT_SIZE), 0.0);
+						tex_posf.xy *= tex_pixel_size;
+
+						vec3 pos_uvw = tex_posf;
+						pos_uvw.xy += vec2(offset.xy) * probe_uv_offset.xy;
+						pos_uvw.x += float(offset.z) * probe_uv_offset.z;
+						vec3 indirect_light = textureLod(sampler2DArray(lightprobe_texture, linear_sampler), pos_uvw, 0.0).rgb;
+
+						light_accum[k] += indirect_light * weight;
+						weight_accum[k] += weight;
+					}
+				}
+			}
+		}
+
+		for (uint k = 0; k < 6; k++) {
+			if (weight_accum[k] > 0.0) {
+				light_accum[k] /= weight_accum[k];
+				light_accum[k] *= feedback;
+			}
+		}
+	}
+
+#endif
+
+	{
+		uint rgbe = process_voxels.data[voxel_index].light;
+
+		//read rgbe8985
+		float r = float((rgbe & 0xff) << 1);
+		float g = float((rgbe >> 8) & 0x1ff);
+		float b = float(((rgbe >> 17) & 0xff) << 1);
+		float e = float((rgbe >> 25) & 0x1F);
+		float m = pow(2.0, e - 15.0 - 9.0);
+
+		vec3 l = vec3(r, g, b) * m;
+
+		uint aniso = process_voxels.data[voxel_index].light_aniso;
+		for (uint i = 0; i < 6; i++) {
+			float strength = ((aniso >> (i * 5)) & 0x1F) / float(0x1F);
+			light_accum[i] += l * strength;
+		}
+	}
+
+	// Raytrace light
+
+	vec3 pos_to_uvw = 1.0 / params.grid_size;
+	vec3 uvw_ofs = pos_to_uvw * 0.5;
+
+	for (uint i = 0; i < params.light_count; i++) {
+		float attenuation = 1.0;
+		vec3 direction;
+		float light_distance = 1e20;
+
+		switch (lights.data[i].type) {
+			case LIGHT_TYPE_DIRECTIONAL: {
+				direction = -lights.data[i].direction;
+			} break;
+			case LIGHT_TYPE_OMNI: {
+				vec3 rel_vec = lights.data[i].position - position;
+				direction = normalize(rel_vec);
+				light_distance = length(rel_vec);
+				rel_vec.y /= params.y_mult;
+				attenuation = get_omni_attenuation(light_distance, 1.0 / lights.data[i].radius, lights.data[i].attenuation);
+
+			} break;
+			case LIGHT_TYPE_SPOT: {
+				vec3 rel_vec = lights.data[i].position - position;
+				direction = normalize(rel_vec);
+				light_distance = length(rel_vec);
+				rel_vec.y /= params.y_mult;
+				attenuation = get_omni_attenuation(light_distance, 1.0 / lights.data[i].radius, lights.data[i].attenuation);
+
+				float cos_spot_angle = lights.data[i].cos_spot_angle;
+				float cos_angle = dot(-direction, lights.data[i].direction);
+
+				if (cos_angle < cos_spot_angle) {
+					continue;
+				}
+
+				float scos = max(cos_angle, cos_spot_angle);
+				float spot_rim = max(0.0001, (1.0 - scos) / (1.0 - cos_spot_angle));
+				attenuation *= 1.0 - pow(spot_rim, lights.data[i].inv_spot_attenuation);
+			} break;
+		}
+
+		if (attenuation < 0.001) {
+			continue;
+		}
+
+		bool hit = false;
+
+		vec3 ray_pos = position;
+		vec3 ray_dir = direction;
+		vec3 inv_dir = 1.0 / ray_dir;
+
+		//this is how to properly bias outgoing rays
+		float cell_size = 1.0 / cascades.data[params.cascade].to_cell;
+		ray_pos += sign(direction) * cell_size * 0.48; // go almost to the box edge but remain inside
+		ray_pos += ray_dir * 0.4 * cell_size; //apply a small bias from there
+
+		for (uint j = params.cascade; j < params.max_cascades; j++) {
+			//convert to local bounds
+			vec3 pos = ray_pos - cascades.data[j].offset;
+			pos *= cascades.data[j].to_cell;
+			float local_distance = light_distance * cascades.data[j].to_cell;
+
+			if (any(lessThan(pos, vec3(0.0))) || any(greaterThanEqual(pos, params.grid_size))) {
+				continue; //already past bounds for this cascade, goto next
+			}
+
+			//find maximum advance distance (until reaching bounds)
+			vec3 t0 = -pos * inv_dir;
+			vec3 t1 = (params.grid_size - pos) * inv_dir;
+			vec3 tmax = max(t0, t1);
+			float max_advance = min(tmax.x, min(tmax.y, tmax.z));
+
+			max_advance = min(local_distance, max_advance);
+
+			float advance = 0.0;
+			float occlusion = 1.0;
+
+			while (advance < max_advance) {
+				//read how much to advance from SDF
+				vec3 uvw = (pos + ray_dir * advance) * pos_to_uvw;
+
+				float distance = texture(sampler3D(sdf_cascades[j], linear_sampler), uvw).r * 255.0 - 1.0;
+				if (distance < 0.001) {
+					//consider hit
+					hit = true;
+					break;
+				}
+
+				occlusion = min(occlusion, distance);
+
+				advance += distance;
+			}
+
+			if (hit) {
+				attenuation *= occlusion;
+				break;
+			}
+
+			if (advance >= local_distance) {
+				break; //past light distance, abandon search
+			}
+			//change ray origin to collision with bounds
+			pos += ray_dir * max_advance;
+			pos /= cascades.data[j].to_cell;
+			pos += cascades.data[j].offset;
+			light_distance -= max_advance / cascades.data[j].to_cell;
+			ray_pos = pos;
+		}
+
+		if (!hit) {
+			vec3 light = albedo * lights.data[i].color.rgb * lights.data[i].energy * attenuation;
+
+			for (int j = 0; j < 6; j++) {
+				if (bool(valid_aniso & (1 << j))) {
+					light_accum[j] += max(0.0, dot(aniso_dir[j], direction)) * light;
+				}
+			}
+		}
+	}
+
+	// Store the light in the light texture
+
+	float lumas[6];
+	vec3 light_total = vec3(0);
+
+	for (int i = 0; i < 6; i++) {
+		light_total += light_accum[i];
+		lumas[i] = max(light_accum[i].r, max(light_accum[i].g, light_accum[i].b));
+	}
+
+	float luma_total = max(light_total.r, max(light_total.g, light_total.b));
+
+	uint light_total_rgbe;
+
+	{
+		//compress to RGBE9995 to save space
+
+		const float pow2to9 = 512.0f;
+		const float B = 15.0f;
+		const float N = 9.0f;
+		const float LN2 = 0.6931471805599453094172321215;
+
+		float cRed = clamp(light_total.r, 0.0, 65408.0);
+		float cGreen = clamp(light_total.g, 0.0, 65408.0);
+		float cBlue = clamp(light_total.b, 0.0, 65408.0);
+
+		float cMax = max(cRed, max(cGreen, cBlue));
+
+		float expp = max(-B - 1.0f, floor(log(cMax) / LN2)) + 1.0f + B;
+
+		float sMax = floor((cMax / pow(2.0f, expp - B - N)) + 0.5f);
+
+		float exps = expp + 1.0f;
+
+		if (0.0 <= sMax && sMax < pow2to9) {
+			exps = expp;
+		}
+
+		float sRed = floor((cRed / pow(2.0f, exps - B - N)) + 0.5f);
+		float sGreen = floor((cGreen / pow(2.0f, exps - B - N)) + 0.5f);
+		float sBlue = floor((cBlue / pow(2.0f, exps - B - N)) + 0.5f);
+#ifdef MODE_PROCESS_STATIC
+		//since its self-save, use RGBE8985
+		light_total_rgbe = ((uint(sRed) & 0x1FF) >> 1) | ((uint(sGreen) & 0x1FF) << 8) | (((uint(sBlue) & 0x1FF) >> 1) << 17) | ((uint(exps) & 0x1F) << 25);
+
+#else
+		light_total_rgbe = (uint(sRed) & 0x1FF) | ((uint(sGreen) & 0x1FF) << 9) | ((uint(sBlue) & 0x1FF) << 18) | ((uint(exps) & 0x1F) << 27);
+#endif
+	}
+
+#ifdef MODE_PROCESS_DYNAMIC
+
+	vec4 aniso0;
+	aniso0.r = lumas[0] / luma_total;
+	aniso0.g = lumas[1] / luma_total;
+	aniso0.b = lumas[2] / luma_total;
+	aniso0.a = lumas[3] / luma_total;
+
+	vec2 aniso1;
+	aniso1.r = lumas[4] / luma_total;
+	aniso1.g = lumas[5] / luma_total;
+
+	//save to 3D textures
+	imageStore(dst_aniso0, positioni, aniso0);
+	imageStore(dst_aniso1, positioni, vec4(aniso1, 0.0, 0.0));
+	imageStore(dst_light, positioni, uvec4(light_total_rgbe));
+
+	//also fill neighbors, so light interpolation during the indirect pass works
+
+	//recover the neighbor list from the leftover bits
+	uint neighbors = (voxel_albedo >> 21) | ((voxel_position >> 21) << 11) | ((process_voxels.data[voxel_index].light >> 30) << 22) | ((process_voxels.data[voxel_index].light_aniso >> 30) << 24);
+
+	const uint max_neighbours = 26;
+	const ivec3 neighbour_positions[max_neighbours] = ivec3[](
+			ivec3(-1, -1, -1),
+			ivec3(-1, -1, 0),
+			ivec3(-1, -1, 1),
+			ivec3(-1, 0, -1),
+			ivec3(-1, 0, 0),
+			ivec3(-1, 0, 1),
+			ivec3(-1, 1, -1),
+			ivec3(-1, 1, 0),
+			ivec3(-1, 1, 1),
+			ivec3(0, -1, -1),
+			ivec3(0, -1, 0),
+			ivec3(0, -1, 1),
+			ivec3(0, 0, -1),
+			ivec3(0, 0, 1),
+			ivec3(0, 1, -1),
+			ivec3(0, 1, 0),
+			ivec3(0, 1, 1),
+			ivec3(1, -1, -1),
+			ivec3(1, -1, 0),
+			ivec3(1, -1, 1),
+			ivec3(1, 0, -1),
+			ivec3(1, 0, 0),
+			ivec3(1, 0, 1),
+			ivec3(1, 1, -1),
+			ivec3(1, 1, 0),
+			ivec3(1, 1, 1));
+
+	for (uint i = 0; i < max_neighbours; i++) {
+		if (bool(neighbors & (1 << i))) {
+			ivec3 neighbour_pos = positioni + neighbour_positions[i];
+			imageStore(dst_light, neighbour_pos, uvec4(light_total_rgbe));
+			imageStore(dst_aniso0, neighbour_pos, aniso0);
+			imageStore(dst_aniso1, neighbour_pos, vec4(aniso1, 0.0, 0.0));
+		}
+	}
+
+#endif
+
+#ifdef MODE_PROCESS_STATIC
+
+	//save back the anisotropic
+
+	uint light = process_voxels.data[voxel_index].light & (3 << 30);
+	light |= light_total_rgbe;
+	process_voxels.data[voxel_index].light = light; //replace
+
+	uint light_aniso = process_voxels.data[voxel_index].light_aniso & (3 << 30);
+	for (int i = 0; i < 6; i++) {
+		light_aniso |= min(31, uint((lumas[i] / luma_total) * 31.0)) << (i * 5);
+	}
+
+	process_voxels.data[voxel_index].light_aniso = light_aniso;
+
+#endif
+}

servers/rendering/renderer_rd/shaders/environment/sdfgi_integrate.glsl

@@ -0,0 +1,618 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+#define MAX_CASCADES 8
+
+layout(set = 0, binding = 1) uniform texture3D sdf_cascades[MAX_CASCADES];
+layout(set = 0, binding = 2) uniform texture3D light_cascades[MAX_CASCADES];
+layout(set = 0, binding = 3) uniform texture3D aniso0_cascades[MAX_CASCADES];
+layout(set = 0, binding = 4) uniform texture3D aniso1_cascades[MAX_CASCADES];
+
+layout(set = 0, binding = 6) uniform sampler linear_sampler;
+
+struct CascadeData {
+	vec3 offset; //offset of (0,0,0) in world coordinates
+	float to_cell; // 1/bounds * grid_size
+	ivec3 probe_world_offset;
+	uint pad;
+	vec4 pad2;
+};
+
+layout(set = 0, binding = 7, std140) uniform Cascades {
+	CascadeData data[MAX_CASCADES];
+}
+cascades;
+
+layout(r32ui, set = 0, binding = 8) uniform restrict uimage2DArray lightprobe_texture_data;
+layout(rgba16i, set = 0, binding = 9) uniform restrict iimage2DArray lightprobe_history_texture;
+layout(rgba32i, set = 0, binding = 10) uniform restrict iimage2D lightprobe_average_texture;
+
+//used for scrolling
+layout(rgba16i, set = 0, binding = 11) uniform restrict iimage2DArray lightprobe_history_scroll_texture;
+layout(rgba32i, set = 0, binding = 12) uniform restrict iimage2D lightprobe_average_scroll_texture;
+
+layout(rgba32i, set = 0, binding = 13) uniform restrict iimage2D lightprobe_average_parent_texture;
+
+layout(rgba16f, set = 0, binding = 14) uniform restrict writeonly image2DArray lightprobe_ambient_texture;
+
+#ifdef USE_CUBEMAP_ARRAY
+layout(set = 1, binding = 0) uniform textureCubeArray sky_irradiance;
+#else
+layout(set = 1, binding = 0) uniform textureCube sky_irradiance;
+#endif
+layout(set = 1, binding = 1) uniform sampler linear_sampler_mipmaps;
+
+#define HISTORY_BITS 10
+
+#define SKY_FLAGS_MODE_COLOR 0x01
+#define SKY_FLAGS_MODE_SKY 0x02
+#define SKY_FLAGS_ORIENTATION_SIGN 0x04
+
+layout(push_constant, std430) uniform Params {
+	vec3 grid_size;
+	uint max_cascades;
+
+	uint probe_axis_size;
+	uint cascade;
+	uint history_index;
+	uint history_size;
+
+	uint ray_count;
+	float ray_bias;
+	ivec2 image_size;
+
+	ivec3 world_offset;
+	uint sky_flags;
+
+	ivec3 scroll;
+	float sky_energy;
+
+	vec3 sky_color_or_orientation;
+	float y_mult;
+
+	bool store_ambient_texture;
+	uint pad[3];
+}
+params;
+
+const float PI = 3.14159265f;
+const float GOLDEN_ANGLE = PI * (3.0 - sqrt(5.0));
+
+vec3 vogel_hemisphere(uint p_index, uint p_count, float p_offset) {
+	float r = sqrt(float(p_index) + 0.5f) / sqrt(float(p_count));
+	float theta = float(p_index) * GOLDEN_ANGLE + p_offset;
+	float y = cos(r * PI * 0.5);
+	float l = sin(r * PI * 0.5);
+	return vec3(l * cos(theta), l * sin(theta), y * (float(p_index & 1) * 2.0 - 1.0));
+}
+
+uvec3 hash3(uvec3 x) {
+	x = ((x >> 16) ^ x) * 0x45d9f3b;
+	x = ((x >> 16) ^ x) * 0x45d9f3b;
+	x = (x >> 16) ^ x;
+	return x;
+}
+
+float hashf3(vec3 co) {
+	return fract(sin(dot(co, vec3(12.9898, 78.233, 137.13451))) * 43758.5453);
+}
+
+vec3 octahedron_encode(vec2 f) {
+	// https://twitter.com/Stubbesaurus/status/937994790553227264
+	f = f * 2.0 - 1.0;
+	vec3 n = vec3(f.x, f.y, 1.0f - abs(f.x) - abs(f.y));
+	float t = clamp(-n.z, 0.0, 1.0);
+	n.x += n.x >= 0 ? -t : t;
+	n.y += n.y >= 0 ? -t : t;
+	return normalize(n);
+}
+
+uint rgbe_encode(vec3 color) {
+	const float pow2to9 = 512.0f;
+	const float B = 15.0f;
+	const float N = 9.0f;
+	const float LN2 = 0.6931471805599453094172321215;
+
+	float cRed = clamp(color.r, 0.0, 65408.0);
+	float cGreen = clamp(color.g, 0.0, 65408.0);
+	float cBlue = clamp(color.b, 0.0, 65408.0);
+
+	float cMax = max(cRed, max(cGreen, cBlue));
+
+	float expp = max(-B - 1.0f, floor(log(cMax) / LN2)) + 1.0f + B;
+
+	float sMax = floor((cMax / pow(2.0f, expp - B - N)) + 0.5f);
+
+	float exps = expp + 1.0f;
+
+	if (0.0 <= sMax && sMax < pow2to9) {
+		exps = expp;
+	}
+
+	float sRed = floor((cRed / pow(2.0f, exps - B - N)) + 0.5f);
+	float sGreen = floor((cGreen / pow(2.0f, exps - B - N)) + 0.5f);
+	float sBlue = floor((cBlue / pow(2.0f, exps - B - N)) + 0.5f);
+	return (uint(sRed) & 0x1FF) | ((uint(sGreen) & 0x1FF) << 9) | ((uint(sBlue) & 0x1FF) << 18) | ((uint(exps) & 0x1F) << 27);
+}
+
+struct SH {
+#if (SH_SIZE == 16)
+	float c[48];
+#else
+	float c[28];
+#endif
+};
+
+shared SH sh_accum[64]; //8x8
+
+void main() {
+	ivec2 pos = ivec2(gl_GlobalInvocationID.xy);
+	if (any(greaterThanEqual(pos, params.image_size))) { //too large, do nothing
+		return;
+	}
+
+	uint probe_index = gl_LocalInvocationID.x + gl_LocalInvocationID.y * 8;
+
+#ifdef MODE_PROCESS
+
+	float probe_cell_size = float(params.grid_size.x / float(params.probe_axis_size - 1)) / cascades.data[params.cascade].to_cell;
+
+	ivec3 probe_cell;
+	probe_cell.x = pos.x % int(params.probe_axis_size);
+	probe_cell.y = pos.y;
+	probe_cell.z = pos.x / int(params.probe_axis_size);
+
+	vec3 probe_pos = cascades.data[params.cascade].offset + vec3(probe_cell) * probe_cell_size;
+	vec3 pos_to_uvw = 1.0 / params.grid_size;
+
+	for (uint i = 0; i < SH_SIZE * 3; i++) {
+		sh_accum[probe_index].c[i] = 0.0;
+	}
+
+	// quickly ensure each probe has a different "offset" for the vogel function, based on integer world position
+	uvec3 h3 = hash3(uvec3(params.world_offset + probe_cell));
+	float offset = hashf3(vec3(h3 & uvec3(0xFFFFF)));
+
+	//for a more homogeneous hemisphere, alternate based on history frames
+	uint ray_offset = params.history_index;
+	uint ray_mult = params.history_size;
+	uint ray_total = ray_mult * params.ray_count;
+
+	for (uint i = 0; i < params.ray_count; i++) {
+		vec3 ray_dir = vogel_hemisphere(ray_offset + i * ray_mult, ray_total, offset);
+		ray_dir.y *= params.y_mult;
+		ray_dir = normalize(ray_dir);
+
+		//needs to be visible
+		vec3 ray_pos = probe_pos;
+		vec3 inv_dir = 1.0 / ray_dir;
+
+		bool hit = false;
+		uint hit_cascade;
+
+		float bias = params.ray_bias;
+		vec3 abs_ray_dir = abs(ray_dir);
+		ray_pos += ray_dir * 1.0 / max(abs_ray_dir.x, max(abs_ray_dir.y, abs_ray_dir.z)) * bias / cascades.data[params.cascade].to_cell;
+		vec3 uvw;
+
+		for (uint j = params.cascade; j < params.max_cascades; j++) {
+			//convert to local bounds
+			vec3 pos = ray_pos - cascades.data[j].offset;
+			pos *= cascades.data[j].to_cell;
+
+			if (any(lessThan(pos, vec3(0.0))) || any(greaterThanEqual(pos, params.grid_size))) {
+				continue; //already past bounds for this cascade, goto next
+			}
+
+			//find maximum advance distance (until reaching bounds)
+			vec3 t0 = -pos * inv_dir;
+			vec3 t1 = (params.grid_size - pos) * inv_dir;
+			vec3 tmax = max(t0, t1);
+			float max_advance = min(tmax.x, min(tmax.y, tmax.z));
+
+			float advance = 0.0;
+
+			while (advance < max_advance) {
+				//read how much to advance from SDF
+				uvw = (pos + ray_dir * advance) * pos_to_uvw;
+
+				float distance = texture(sampler3D(sdf_cascades[j], linear_sampler), uvw).r * 255.0 - 1.0;
+				if (distance < 0.05) {
+					//consider hit
+					hit = true;
+					break;
+				}
+
+				advance += distance;
+			}
+
+			if (hit) {
+				hit_cascade = j;
+				break;
+			}
+
+			//change ray origin to collision with bounds
+			pos += ray_dir * max_advance;
+			pos /= cascades.data[j].to_cell;
+			pos += cascades.data[j].offset;
+			ray_pos = pos;
+		}
+
+		vec4 light;
+		if (hit) {
+			//avoid reading different texture from different threads
+			for (uint j = params.cascade; j < params.max_cascades; j++) {
+				if (j == hit_cascade) {
+					const float EPSILON = 0.001;
+					vec3 hit_normal = normalize(vec3(
+							texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw + vec3(EPSILON, 0.0, 0.0)).r - texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw - vec3(EPSILON, 0.0, 0.0)).r,
+							texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw + vec3(0.0, EPSILON, 0.0)).r - texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw - vec3(0.0, EPSILON, 0.0)).r,
+							texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw + vec3(0.0, 0.0, EPSILON)).r - texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw - vec3(0.0, 0.0, EPSILON)).r));
+
+					vec3 hit_light = texture(sampler3D(light_cascades[hit_cascade], linear_sampler), uvw).rgb;
+					vec4 aniso0 = texture(sampler3D(aniso0_cascades[hit_cascade], linear_sampler), uvw);
+					vec3 hit_aniso0 = aniso0.rgb;
+					vec3 hit_aniso1 = vec3(aniso0.a, texture(sampler3D(aniso1_cascades[hit_cascade], linear_sampler), uvw).rg);
+
+					//one liner magic
+					light.rgb = hit_light * (dot(max(vec3(0.0), (hit_normal * hit_aniso0)), vec3(1.0)) + dot(max(vec3(0.0), (-hit_normal * hit_aniso1)), vec3(1.0)));
+					light.a = 1.0;
+				}
+			}
+
+		} else if (bool(params.sky_flags & SKY_FLAGS_MODE_SKY)) {
+			// Reconstruct sky orientation as quaternion and rotate ray_dir before sampling.
+			float sky_sign = bool(params.sky_flags & SKY_FLAGS_ORIENTATION_SIGN) ? 1.0 : -1.0;
+			vec4 sky_quat = vec4(params.sky_color_or_orientation, sky_sign * sqrt(1.0 - dot(params.sky_color_or_orientation, params.sky_color_or_orientation)));
+			vec3 sky_dir = cross(sky_quat.xyz, ray_dir);
+			sky_dir = ray_dir + ((sky_dir * sky_quat.w) + cross(sky_quat.xyz, sky_dir)) * 2.0;
+#ifdef USE_CUBEMAP_ARRAY
+			light.rgb = textureLod(samplerCubeArray(sky_irradiance, linear_sampler_mipmaps), vec4(sky_dir, 0.0), 2.0).rgb; // Use second mipmap because we don't usually throw a lot of rays, so this compensates.
+#else
+			light.rgb = textureLod(samplerCube(sky_irradiance, linear_sampler_mipmaps), sky_dir, 2.0).rgb; // Use second mipmap because we don't usually throw a lot of rays, so this compensates.
+#endif
+			light.rgb *= params.sky_energy;
+			light.a = 0.0;
+
+		} else if (bool(params.sky_flags & SKY_FLAGS_MODE_COLOR)) {
+			light.rgb = params.sky_color_or_orientation;
+			light.rgb *= params.sky_energy;
+			light.a = 0.0;
+		} else {
+			light = vec4(0, 0, 0, 0);
+		}
+
+		vec3 ray_dir2 = ray_dir * ray_dir;
+
+#define SH_ACCUM(m_idx, m_value)                       \
+	{                                                  \
+		vec3 l = light.rgb * (m_value);                \
+		sh_accum[probe_index].c[m_idx * 3 + 0] += l.r; \
+		sh_accum[probe_index].c[m_idx * 3 + 1] += l.g; \
+		sh_accum[probe_index].c[m_idx * 3 + 2] += l.b; \
+	}
+		SH_ACCUM(0, 0.282095); //l0
+		SH_ACCUM(1, 0.488603 * ray_dir.y); //l1n1
+		SH_ACCUM(2, 0.488603 * ray_dir.z); //l1n0
+		SH_ACCUM(3, 0.488603 * ray_dir.x); //l1p1
+		SH_ACCUM(4, 1.092548 * ray_dir.x * ray_dir.y); //l2n2
+		SH_ACCUM(5, 1.092548 * ray_dir.y * ray_dir.z); //l2n1
+		SH_ACCUM(6, 0.315392 * (3.0 * ray_dir2.z - 1.0)); //l20
+		SH_ACCUM(7, 1.092548 * ray_dir.x * ray_dir.z); //l2p1
+		SH_ACCUM(8, 0.546274 * (ray_dir2.x - ray_dir2.y)); //l2p2
+#if (SH_SIZE == 16)
+		SH_ACCUM(9, 0.590043 * ray_dir.y * (3.0f * ray_dir2.x - ray_dir2.y));
+		SH_ACCUM(10, 2.890611 * ray_dir.y * ray_dir.x * ray_dir.z);
+		SH_ACCUM(11, 0.646360 * ray_dir.y * (-1.0f + 5.0f * ray_dir2.z));
+		SH_ACCUM(12, 0.373176 * (5.0f * ray_dir2.z * ray_dir.z - 3.0f * ray_dir.z));
+		SH_ACCUM(13, 0.457045 * ray_dir.x * (-1.0f + 5.0f * ray_dir2.z));
+		SH_ACCUM(14, 1.445305 * (ray_dir2.x - ray_dir2.y) * ray_dir.z);
+		SH_ACCUM(15, 0.590043 * ray_dir.x * (ray_dir2.x - 3.0f * ray_dir2.y));
+
+#endif
+	}
+
+	for (uint i = 0; i < SH_SIZE; i++) {
+		// store in history texture
+		ivec3 prev_pos = ivec3(pos.x, pos.y * SH_SIZE + i, int(params.history_index));
+		ivec2 average_pos = prev_pos.xy;
+
+		vec4 value = vec4(sh_accum[probe_index].c[i * 3 + 0], sh_accum[probe_index].c[i * 3 + 1], sh_accum[probe_index].c[i * 3 + 2], 1.0) * 4.0 / float(params.ray_count);
+
+		ivec4 ivalue = clamp(ivec4(value * float(1 << HISTORY_BITS)), -32768, 32767); //clamp to 16 bits, so higher values don't break average
+
+		ivec4 prev_value = imageLoad(lightprobe_history_texture, prev_pos);
+		ivec4 average = imageLoad(lightprobe_average_texture, average_pos);
+
+		average -= prev_value;
+		average += ivalue;
+
+		imageStore(lightprobe_history_texture, prev_pos, ivalue);
+		imageStore(lightprobe_average_texture, average_pos, average);
+
+		if (params.store_ambient_texture && i == 0) {
+			ivec3 ambient_pos = ivec3(pos, int(params.cascade));
+			vec4 ambient_light = (vec4(average) / float(params.history_size)) / float(1 << HISTORY_BITS);
+			ambient_light *= 0.88622; // SHL0
+			imageStore(lightprobe_ambient_texture, ambient_pos, ambient_light);
+		}
+	}
+#endif // MODE PROCESS
+
+#ifdef MODE_STORE
+
+	// converting to octahedral in this step is required because
+	// octahedral is much faster to read from the screen than spherical harmonics,
+	// despite the very slight quality loss
+
+	ivec2 sh_pos = (pos / OCT_SIZE) * ivec2(1, SH_SIZE);
+	ivec2 oct_pos = (pos / OCT_SIZE) * (OCT_SIZE + 2) + ivec2(1);
+	ivec2 local_pos = pos % OCT_SIZE;
+
+	//compute the octahedral normal for this texel
+	vec3 normal = octahedron_encode(vec2(local_pos) / float(OCT_SIZE));
+
+	// read the spherical harmonic
+
+	vec3 normal2 = normal * normal;
+	float c[SH_SIZE] = float[](
+
+			0.282095, //l0
+			0.488603 * normal.y, //l1n1
+			0.488603 * normal.z, //l1n0
+			0.488603 * normal.x, //l1p1
+			1.092548 * normal.x * normal.y, //l2n2
+			1.092548 * normal.y * normal.z, //l2n1
+			0.315392 * (3.0 * normal2.z - 1.0), //l20
+			1.092548 * normal.x * normal.z, //l2p1
+			0.546274 * (normal2.x - normal2.y) //l2p2
+#if (SH_SIZE == 16)
+			,
+			0.590043 * normal.y * (3.0f * normal2.x - normal2.y),
+			2.890611 * normal.y * normal.x * normal.z,
+			0.646360 * normal.y * (-1.0f + 5.0f * normal2.z),
+			0.373176 * (5.0f * normal2.z * normal.z - 3.0f * normal.z),
+			0.457045 * normal.x * (-1.0f + 5.0f * normal2.z),
+			1.445305 * (normal2.x - normal2.y) * normal.z,
+			0.590043 * normal.x * (normal2.x - 3.0f * normal2.y)
+
+#endif
+	);
+
+	const float l_mult[SH_SIZE] = float[](
+			1.0,
+			2.0 / 3.0,
+			2.0 / 3.0,
+			2.0 / 3.0,
+			1.0 / 4.0,
+			1.0 / 4.0,
+			1.0 / 4.0,
+			1.0 / 4.0,
+			1.0 / 4.0
+#if (SH_SIZE == 16)
+			, // l4 does not contribute to irradiance
+			0.0,
+			0.0,
+			0.0,
+			0.0,
+			0.0,
+			0.0,
+			0.0
+#endif
+	);
+
+	vec3 irradiance = vec3(0.0);
+	vec3 radiance = vec3(0.0);
+
+	for (uint i = 0; i < SH_SIZE; i++) {
+		// store in history texture
+		ivec2 average_pos = sh_pos + ivec2(0, i);
+		ivec4 average = imageLoad(lightprobe_average_texture, average_pos);
+
+		vec4 sh = (vec4(average) / float(params.history_size)) / float(1 << HISTORY_BITS);
+
+		vec3 m = sh.rgb * c[i] * 4.0;
+
+		irradiance += m * l_mult[i];
+		radiance += m;
+	}
+
+	//encode RGBE9995 for the final texture
+
+	uint irradiance_rgbe = rgbe_encode(irradiance);
+	uint radiance_rgbe = rgbe_encode(radiance);
+
+	//store in octahedral map
+
+	ivec3 texture_pos = ivec3(oct_pos, int(params.cascade));
+	ivec3 copy_to[4] = ivec3[](ivec3(-2, -2, -2), ivec3(-2, -2, -2), ivec3(-2, -2, -2), ivec3(-2, -2, -2));
+	copy_to[0] = texture_pos + ivec3(local_pos, 0);
+
+	if (local_pos == ivec2(0, 0)) {
+		copy_to[1] = texture_pos + ivec3(OCT_SIZE - 1, -1, 0);
+		copy_to[2] = texture_pos + ivec3(-1, OCT_SIZE - 1, 0);
+		copy_to[3] = texture_pos + ivec3(OCT_SIZE, OCT_SIZE, 0);
+	} else if (local_pos == ivec2(OCT_SIZE - 1, 0)) {
+		copy_to[1] = texture_pos + ivec3(0, -1, 0);
+		copy_to[2] = texture_pos + ivec3(OCT_SIZE, OCT_SIZE - 1, 0);
+		copy_to[3] = texture_pos + ivec3(-1, OCT_SIZE, 0);
+	} else if (local_pos == ivec2(0, OCT_SIZE - 1)) {
+		copy_to[1] = texture_pos + ivec3(-1, 0, 0);
+		copy_to[2] = texture_pos + ivec3(OCT_SIZE - 1, OCT_SIZE, 0);
+		copy_to[3] = texture_pos + ivec3(OCT_SIZE, -1, 0);
+	} else if (local_pos == ivec2(OCT_SIZE - 1, OCT_SIZE - 1)) {
+		copy_to[1] = texture_pos + ivec3(0, OCT_SIZE, 0);
+		copy_to[2] = texture_pos + ivec3(OCT_SIZE, 0, 0);
+		copy_to[3] = texture_pos + ivec3(-1, -1, 0);
+	} else if (local_pos.y == 0) {
+		copy_to[1] = texture_pos + ivec3(OCT_SIZE - local_pos.x - 1, local_pos.y - 1, 0);
+	} else if (local_pos.x == 0) {
+		copy_to[1] = texture_pos + ivec3(local_pos.x - 1, OCT_SIZE - local_pos.y - 1, 0);
+	} else if (local_pos.y == OCT_SIZE - 1) {
+		copy_to[1] = texture_pos + ivec3(OCT_SIZE - local_pos.x - 1, local_pos.y + 1, 0);
+	} else if (local_pos.x == OCT_SIZE - 1) {
+		copy_to[1] = texture_pos + ivec3(local_pos.x + 1, OCT_SIZE - local_pos.y - 1, 0);
+	}
+
+	for (int i = 0; i < 4; i++) {
+		if (copy_to[i] == ivec3(-2, -2, -2)) {
+			continue;
+		}
+		imageStore(lightprobe_texture_data, copy_to[i], uvec4(irradiance_rgbe));
+		imageStore(lightprobe_texture_data, copy_to[i] + ivec3(0, 0, int(params.max_cascades)), uvec4(radiance_rgbe));
+	}
+
+#endif
+
+#ifdef MODE_SCROLL
+
+	ivec3 probe_cell;
+	probe_cell.x = pos.x % int(params.probe_axis_size);
+	probe_cell.y = pos.y;
+	probe_cell.z = pos.x / int(params.probe_axis_size);
+
+	ivec3 read_probe = probe_cell - params.scroll;
+
+	if (all(greaterThanEqual(read_probe, ivec3(0))) && all(lessThan(read_probe, ivec3(params.probe_axis_size)))) {
+		// can scroll
+		ivec2 tex_pos;
+		tex_pos = read_probe.xy;
+		tex_pos.x += read_probe.z * int(params.probe_axis_size);
+
+		//scroll
+		for (uint j = 0; j < params.history_size; j++) {
+			for (int i = 0; i < SH_SIZE; i++) {
+				// copy from history texture
+				ivec3 src_pos = ivec3(tex_pos.x, tex_pos.y * SH_SIZE + i, int(j));
+				ivec3 dst_pos = ivec3(pos.x, pos.y * SH_SIZE + i, int(j));
+				ivec4 value = imageLoad(lightprobe_history_texture, src_pos);
+				imageStore(lightprobe_history_scroll_texture, dst_pos, value);
+			}
+		}
+
+		for (int i = 0; i < SH_SIZE; i++) {
+			// copy from average texture
+			ivec2 src_pos = ivec2(tex_pos.x, tex_pos.y * SH_SIZE + i);
+			ivec2 dst_pos = ivec2(pos.x, pos.y * SH_SIZE + i);
+			ivec4 value = imageLoad(lightprobe_average_texture, src_pos);
+			imageStore(lightprobe_average_scroll_texture, dst_pos, value);
+		}
+	} else if (params.cascade < params.max_cascades - 1) {
+		//can't scroll, must look for position in parent cascade
+
+		//to global coords
+		float cell_to_probe = float(params.grid_size.x / float(params.probe_axis_size - 1));
+
+		float probe_cell_size = cell_to_probe / cascades.data[params.cascade].to_cell;
+		vec3 probe_pos = cascades.data[params.cascade].offset + vec3(probe_cell) * probe_cell_size;
+
+		//to parent local coords
+		float probe_cell_size_next = cell_to_probe / cascades.data[params.cascade + 1].to_cell;
+		probe_pos -= cascades.data[params.cascade + 1].offset;
+		probe_pos /= probe_cell_size_next;
+
+		ivec3 probe_posi = ivec3(probe_pos);
+		//add up all light, no need to use occlusion here, since occlusion will do its work afterwards
+
+		vec4 average_light[SH_SIZE] = vec4[](vec4(0), vec4(0), vec4(0), vec4(0), vec4(0), vec4(0), vec4(0), vec4(0), vec4(0)
+#if (SH_SIZE == 16)
+																															 ,
+				vec4(0), vec4(0), vec4(0), vec4(0), vec4(0), vec4(0), vec4(0)
+#endif
+		);
+		float total_weight = 0.0;
+
+		for (int i = 0; i < 8; i++) {
+			ivec3 offset = probe_posi + ((ivec3(i) >> ivec3(0, 1, 2)) & ivec3(1, 1, 1));
+
+			vec3 trilinear = vec3(1.0) - abs(probe_pos - vec3(offset));
+			float weight = trilinear.x * trilinear.y * trilinear.z;
+
+			ivec2 tex_pos;
+			tex_pos = offset.xy;
+			tex_pos.x += offset.z * int(params.probe_axis_size);
+
+			for (int j = 0; j < SH_SIZE; j++) {
+				// copy from history texture
+				ivec2 src_pos = ivec2(tex_pos.x, tex_pos.y * SH_SIZE + j);
+				ivec4 average = imageLoad(lightprobe_average_parent_texture, src_pos);
+				vec4 value = (vec4(average) / float(params.history_size)) / float(1 << HISTORY_BITS);
+				average_light[j] += value * weight;
+			}
+
+			total_weight += weight;
+		}
+
+		if (total_weight > 0.0) {
+			total_weight = 1.0 / total_weight;
+		}
+		//store the averaged values everywhere
+
+		for (int i = 0; i < SH_SIZE; i++) {
+			ivec4 ivalue = clamp(ivec4(average_light[i] * total_weight * float(1 << HISTORY_BITS)), ivec4(-32768), ivec4(32767)); //clamp to 16 bits, so higher values don't break average
+			// copy from history texture
+			ivec3 dst_pos = ivec3(pos.x, pos.y * SH_SIZE + i, 0);
+			for (uint j = 0; j < params.history_size; j++) {
+				dst_pos.z = int(j);
+				imageStore(lightprobe_history_scroll_texture, dst_pos, ivalue);
+			}
+
+			ivalue *= int(params.history_size); //average needs to have all history added up
+			imageStore(lightprobe_average_scroll_texture, dst_pos.xy, ivalue);
+		}
+
+	} else {
+		//scroll at the edge of the highest cascade, just copy what is there,
+		//since its the closest we have anyway
+
+		for (uint j = 0; j < params.history_size; j++) {
+			ivec2 tex_pos;
+			tex_pos = probe_cell.xy;
+			tex_pos.x += probe_cell.z * int(params.probe_axis_size);
+
+			for (int i = 0; i < SH_SIZE; i++) {
+				// copy from history texture
+				ivec3 src_pos = ivec3(tex_pos.x, tex_pos.y * SH_SIZE + i, int(j));
+				ivec3 dst_pos = ivec3(pos.x, pos.y * SH_SIZE + i, int(j));
+				ivec4 value = imageLoad(lightprobe_history_texture, dst_pos);
+				imageStore(lightprobe_history_scroll_texture, dst_pos, value);
+			}
+		}
+
+		for (int i = 0; i < SH_SIZE; i++) {
+			// copy from average texture
+			ivec2 spos = ivec2(pos.x, pos.y * SH_SIZE + i);
+			ivec4 average = imageLoad(lightprobe_average_texture, spos);
+			imageStore(lightprobe_average_scroll_texture, spos, average);
+		}
+	}
+
+#endif
+
+#ifdef MODE_SCROLL_STORE
+
+	//do not update probe texture, as these will be updated later
+
+	for (uint j = 0; j < params.history_size; j++) {
+		for (int i = 0; i < SH_SIZE; i++) {
+			// copy from history texture
+			ivec3 spos = ivec3(pos.x, pos.y * SH_SIZE + i, int(j));
+			ivec4 value = imageLoad(lightprobe_history_scroll_texture, spos);
+			imageStore(lightprobe_history_texture, spos, value);
+		}
+	}
+
+	for (int i = 0; i < SH_SIZE; i++) {
+		// copy from average texture
+		ivec2 spos = ivec2(pos.x, pos.y * SH_SIZE + i);
+		ivec4 average = imageLoad(lightprobe_average_scroll_texture, spos);
+		imageStore(lightprobe_average_texture, spos, average);
+	}
+
+#endif
+}

servers/rendering/renderer_rd/shaders/environment/sky.glsl

@@ -0,0 +1,306 @@
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(location = 0) out vec2 uv_interp;
+
+layout(push_constant, std430) uniform Params {
+	mat3 orientation;
+	vec4 projection; // only applicable if not multiview
+	vec3 position;
+	float time;
+	vec2 pad;
+	float luminance_multiplier;
+	float brightness_multiplier;
+}
+params;
+
+void main() {
+	vec2 base_arr[3] = vec2[](vec2(-1.0, -3.0), vec2(-1.0, 1.0), vec2(3.0, 1.0));
+	uv_interp = base_arr[gl_VertexIndex];
+	gl_Position = vec4(uv_interp, 0.0, 1.0);
+}
+
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+#ifdef USE_MULTIVIEW
+#extension GL_EXT_multiview : enable
+#define ViewIndex gl_ViewIndex
+#endif
+
+#define M_PI 3.14159265359
+
+layout(location = 0) in vec2 uv_interp;
+
+layout(push_constant, std430) uniform Params {
+	mat3 orientation;
+	vec4 projection; // only applicable if not multiview
+	vec3 position;
+	float time;
+	vec2 pad;
+	float luminance_multiplier;
+	float brightness_multiplier;
+}
+params;
+
+#include "../samplers_inc.glsl"
+
+layout(set = 0, binding = 1, std430) restrict readonly buffer GlobalShaderUniformData {
+	vec4 data[];
+}
+global_shader_uniforms;
+
+layout(set = 0, binding = 2, std140) uniform SkySceneData {
+	mat4 combined_reprojection[2];
+	mat4 view_inv_projections[2];
+	vec4 view_eye_offsets[2];
+
+	bool volumetric_fog_enabled; // 4 - 4
+	float volumetric_fog_inv_length; // 4 - 8
+	float volumetric_fog_detail_spread; // 4 - 12
+	float volumetric_fog_sky_affect; // 4 - 16
+
+	bool fog_enabled; // 4 - 20
+	float fog_sky_affect; // 4 - 24
+	float fog_density; // 4 - 28
+	float fog_sun_scatter; // 4 - 32
+
+	vec3 fog_light_color; // 12 - 44
+	float fog_aerial_perspective; // 4 - 48
+
+	float z_far; // 4 - 52
+	uint directional_light_count; // 4 - 56
+	uint pad1; // 4 - 60
+	uint pad2; // 4 - 64
+}
+sky_scene_data;
+
+struct DirectionalLightData {
+	vec4 direction_energy;
+	vec4 color_size;
+	bool enabled;
+};
+
+layout(set = 0, binding = 3, std140) uniform DirectionalLights {
+	DirectionalLightData data[MAX_DIRECTIONAL_LIGHT_DATA_STRUCTS];
+}
+directional_lights;
+
+#ifdef MATERIAL_UNIFORMS_USED
+/* clang-format off */
+layout(set = 1, binding = 0, std140) uniform MaterialUniforms {
+#MATERIAL_UNIFORMS
+} material;
+/* clang-format on */
+#endif
+
+layout(set = 2, binding = 0) uniform textureCube radiance;
+#ifdef USE_CUBEMAP_PASS
+layout(set = 2, binding = 1) uniform textureCube half_res;
+layout(set = 2, binding = 2) uniform textureCube quarter_res;
+#elif defined(USE_MULTIVIEW)
+layout(set = 2, binding = 1) uniform texture2DArray half_res;
+layout(set = 2, binding = 2) uniform texture2DArray quarter_res;
+#else
+layout(set = 2, binding = 1) uniform texture2D half_res;
+layout(set = 2, binding = 2) uniform texture2D quarter_res;
+#endif
+
+layout(set = 3, binding = 0) uniform texture3D volumetric_fog_texture;
+
+#ifdef USE_CUBEMAP_PASS
+#define AT_CUBEMAP_PASS true
+#else
+#define AT_CUBEMAP_PASS false
+#endif
+
+#ifdef USE_HALF_RES_PASS
+#define AT_HALF_RES_PASS true
+#else
+#define AT_HALF_RES_PASS false
+#endif
+
+#ifdef USE_QUARTER_RES_PASS
+#define AT_QUARTER_RES_PASS true
+#else
+#define AT_QUARTER_RES_PASS false
+#endif
+
+#GLOBALS
+
+layout(location = 0) out vec4 frag_color;
+
+#ifdef USE_DEBANDING
+// https://www.iryoku.com/next-generation-post-processing-in-call-of-duty-advanced-warfare
+vec3 interleaved_gradient_noise(vec2 pos) {
+	const vec3 magic = vec3(0.06711056f, 0.00583715f, 52.9829189f);
+	float res = fract(magic.z * fract(dot(pos, magic.xy))) * 2.0 - 1.0;
+	return vec3(res, -res, res) / 255.0;
+}
+#endif
+
+vec4 volumetric_fog_process(vec2 screen_uv) {
+#ifdef USE_MULTIVIEW
+	vec4 reprojected = sky_scene_data.combined_reprojection[ViewIndex] * vec4(screen_uv * 2.0 - 1.0, 0.0, 1.0); // Unproject at the far plane
+	vec3 fog_pos = vec3(reprojected.xy / reprojected.w, 1.0) * 0.5 + 0.5;
+#else
+	vec3 fog_pos = vec3(screen_uv, 1.0);
+#endif
+
+	return texture(sampler3D(volumetric_fog_texture, SAMPLER_LINEAR_CLAMP), fog_pos);
+}
+
+vec4 fog_process(vec3 view, vec3 sky_color) {
+	vec3 fog_color = mix(sky_scene_data.fog_light_color, sky_color, sky_scene_data.fog_aerial_perspective);
+
+	if (sky_scene_data.fog_sun_scatter > 0.001) {
+		vec4 sun_scatter = vec4(0.0);
+		float sun_total = 0.0;
+		for (uint i = 0; i < sky_scene_data.directional_light_count; i++) {
+			vec3 light_color = directional_lights.data[i].color_size.xyz * directional_lights.data[i].direction_energy.w;
+			float light_amount = pow(max(dot(view, directional_lights.data[i].direction_energy.xyz), 0.0), 8.0) * M_PI;
+			fog_color += light_color * light_amount * sky_scene_data.fog_sun_scatter;
+		}
+	}
+
+	return vec4(fog_color, 1.0);
+}
+
+// Eberly approximation from https://seblagarde.wordpress.com/2014/12/01/inverse-trigonometric-functions-gpu-optimization-for-amd-gcn-architecture/.
+// input [-1, 1] and output [0, PI]
+float acos_approx(float p_x) {
+	float x = abs(p_x);
+	float res = -0.156583f * x + (M_PI / 2.0);
+	res *= sqrt(1.0f - x);
+	return (p_x >= 0.0) ? res : M_PI - res;
+}
+
+// Based on https://math.stackexchange.com/questions/1098487/atan2-faster-approximation
+// but using the Eberly coefficients from https://seblagarde.wordpress.com/2014/12/01/inverse-trigonometric-functions-gpu-optimization-for-amd-gcn-architecture/.
+float atan2_approx(float y, float x) {
+	float a = min(abs(x), abs(y)) / max(abs(x), abs(y));
+	float s = a * a;
+	float poly = 0.0872929f;
+	poly = -0.301895f + poly * s;
+	poly = 1.0f + poly * s;
+	poly = poly * a;
+
+	float r = abs(y) > abs(x) ? (M_PI / 2.0) - poly : poly;
+	r = x < 0.0 ? M_PI - r : r;
+	r = y < 0.0 ? -r : r;
+
+	return r;
+}
+
+void main() {
+	vec3 cube_normal;
+#ifdef USE_MULTIVIEW
+	// In multiview our projection matrices will contain positional and rotational offsets that we need to properly unproject.
+	vec4 unproject = vec4(uv_interp.x, uv_interp.y, 0.0, 1.0); // unproject at the far plane
+	vec4 unprojected = sky_scene_data.view_inv_projections[ViewIndex] * unproject;
+	cube_normal = unprojected.xyz / unprojected.w;
+
+	// Unproject will give us the position between the eyes, need to re-offset
+	cube_normal += sky_scene_data.view_eye_offsets[ViewIndex].xyz;
+#else
+	cube_normal.z = -1.0;
+	cube_normal.x = (cube_normal.z * (-uv_interp.x - params.projection.x)) / params.projection.y;
+	cube_normal.y = -(cube_normal.z * (uv_interp.y - params.projection.z)) / params.projection.w;
+#endif
+	cube_normal = mat3(params.orientation) * cube_normal;
+	cube_normal = normalize(cube_normal);
+
+	vec2 uv = uv_interp * 0.5 + 0.5;
+
+	vec2 panorama_coords = vec2(atan2_approx(cube_normal.x, -cube_normal.z), acos_approx(cube_normal.y));
+
+	if (panorama_coords.x < 0.0) {
+		panorama_coords.x += M_PI * 2.0;
+	}
+
+	panorama_coords /= vec2(M_PI * 2.0, M_PI);
+
+	vec3 color = vec3(0.0, 0.0, 0.0);
+	float alpha = 1.0; // Only available to subpasses
+	vec4 half_res_color = vec4(1.0);
+	vec4 quarter_res_color = vec4(1.0);
+	vec4 custom_fog = vec4(0.0);
+
+#ifdef USE_CUBEMAP_PASS
+
+#ifdef USES_HALF_RES_COLOR
+	half_res_color = texture(samplerCube(half_res, SAMPLER_LINEAR_WITH_MIPMAPS_CLAMP), cube_normal) / params.luminance_multiplier;
+#endif
+#ifdef USES_QUARTER_RES_COLOR
+	quarter_res_color = texture(samplerCube(quarter_res, SAMPLER_LINEAR_WITH_MIPMAPS_CLAMP), cube_normal) / params.luminance_multiplier;
+#endif
+
+#else
+
+#ifdef USES_HALF_RES_COLOR
+#ifdef USE_MULTIVIEW
+	half_res_color = textureLod(sampler2DArray(half_res, SAMPLER_LINEAR_CLAMP), vec3(uv, ViewIndex), 0.0) / params.luminance_multiplier;
+#else
+	half_res_color = textureLod(sampler2D(half_res, SAMPLER_LINEAR_CLAMP), uv, 0.0) / params.luminance_multiplier;
+#endif // USE_MULTIVIEW
+#endif // USES_HALF_RES_COLOR
+
+#ifdef USES_QUARTER_RES_COLOR
+#ifdef USE_MULTIVIEW
+	quarter_res_color = textureLod(sampler2DArray(quarter_res, SAMPLER_LINEAR_CLAMP), vec3(uv, ViewIndex), 0.0) / params.luminance_multiplier;
+#else
+	quarter_res_color = textureLod(sampler2D(quarter_res, SAMPLER_LINEAR_CLAMP), uv, 0.0) / params.luminance_multiplier;
+#endif // USE_MULTIVIEW
+#endif // USES_QUARTER_RES_COLOR
+
+#endif //USE_CUBEMAP_PASS
+
+	{
+#CODE : SKY
+	}
+
+	frag_color.rgb = color;
+	frag_color.a = alpha;
+
+	// Apply environment 'brightness' setting separately before fog to ensure consistent luminance.
+	frag_color.rgb = frag_color.rgb * params.brightness_multiplier;
+
+#if !defined(DISABLE_FOG) && !defined(USE_CUBEMAP_PASS)
+
+	// Draw "fixed" fog before volumetric fog to ensure volumetric fog can appear in front of the sky.
+	if (sky_scene_data.fog_enabled) {
+		vec4 fog = fog_process(cube_normal, frag_color.rgb);
+		frag_color.rgb = mix(frag_color.rgb, fog.rgb, fog.a * sky_scene_data.fog_sky_affect);
+	}
+
+	if (sky_scene_data.volumetric_fog_enabled) {
+		vec4 fog = volumetric_fog_process(uv);
+		frag_color.rgb = mix(frag_color.rgb, fog.rgb, fog.a * sky_scene_data.volumetric_fog_sky_affect);
+	}
+
+	if (custom_fog.a > 0.0) {
+		frag_color.rgb = mix(frag_color.rgb, custom_fog.rgb, custom_fog.a);
+	}
+
+#endif // DISABLE_FOG
+
+	// For mobile renderer we're multiplying by 0.5 as we're using a UNORM buffer.
+	// For both mobile and clustered, we also bake in the exposure value for the environment and camera.
+	frag_color.rgb = frag_color.rgb * params.luminance_multiplier;
+
+	// Blending is disabled for Sky, so alpha doesn't blend.
+	// Alpha is used for subsurface scattering so make sure it doesn't get applied to Sky.
+	if (!AT_CUBEMAP_PASS && !AT_HALF_RES_PASS && !AT_QUARTER_RES_PASS) {
+		frag_color.a = 0.0;
+	}
+
+#ifdef USE_DEBANDING
+	frag_color.rgb += interleaved_gradient_noise(gl_FragCoord.xy) * params.luminance_multiplier;
+#endif
+}

servers/rendering/renderer_rd/shaders/environment/volumetric_fog.glsl

@@ -0,0 +1,303 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#ifdef USE_VULKAN_MEMORY_MODEL
+#pragma use_vulkan_memory_model
+#endif
+
+layout(local_size_x = 4, local_size_y = 4, local_size_z = 4) in;
+
+#define DENSITY_SCALE 1024.0
+
+#include "../cluster_data_inc.glsl"
+#include "../light_data_inc.glsl"
+
+#define M_PI 3.14159265359
+
+#include "../samplers_inc.glsl"
+
+layout(set = 0, binding = 2, std430) restrict readonly buffer GlobalShaderUniformData {
+	vec4 data[];
+}
+global_shader_uniforms;
+
+layout(push_constant, std430) uniform Params {
+	vec3 position;
+	float pad;
+
+	vec3 size;
+	float pad2;
+
+	ivec3 corner;
+	uint shape;
+
+	mat4 transform;
+}
+params;
+
+#ifdef NO_IMAGE_ATOMICS
+layout(set = 1, binding = 1) volatile buffer emissive_only_map_buffer {
+	uint emissive_only_map[];
+};
+#else
+layout(r32ui, set = 1, binding = 1) uniform volatile uimage3D emissive_only_map;
+#endif
+
+layout(set = 1, binding = 2, std140) uniform SceneParams {
+	vec2 fog_frustum_size_begin;
+	vec2 fog_frustum_size_end;
+
+	float fog_frustum_end;
+	float z_near; //
+	float z_far; //
+	float time;
+
+	ivec3 fog_volume_size;
+	uint directional_light_count; //
+
+	bool use_temporal_reprojection;
+	uint temporal_frame;
+	float detail_spread;
+	float temporal_blend;
+
+	mat4 to_prev_view;
+	mat4 transform;
+}
+scene_params;
+
+#ifdef NO_IMAGE_ATOMICS
+layout(set = 1, binding = 3) volatile buffer density_only_map_buffer {
+	uint density_only_map[];
+};
+layout(set = 1, binding = 4) volatile buffer light_only_map_buffer {
+	uint light_only_map[];
+};
+#else
+layout(r32ui, set = 1, binding = 3) uniform volatile uimage3D density_only_map;
+layout(r32ui, set = 1, binding = 4) uniform volatile uimage3D light_only_map;
+#endif
+
+#ifdef MATERIAL_UNIFORMS_USED
+/* clang-format off */
+layout(set = 2, binding = 0, std140) uniform MaterialUniforms {
+#MATERIAL_UNIFORMS
+} material;
+/* clang-format on */
+#endif
+
+#GLOBALS
+
+float get_depth_at_pos(float cell_depth_size, int z) {
+	float d = float(z) * cell_depth_size + cell_depth_size * 0.5; //center of voxels
+	d = pow(d, scene_params.detail_spread);
+	return scene_params.fog_frustum_end * d;
+}
+
+#define TEMPORAL_FRAMES 16
+
+const vec3 halton_map[TEMPORAL_FRAMES] = vec3[](
+		vec3(0.5, 0.33333333, 0.2),
+		vec3(0.25, 0.66666667, 0.4),
+		vec3(0.75, 0.11111111, 0.6),
+		vec3(0.125, 0.44444444, 0.8),
+		vec3(0.625, 0.77777778, 0.04),
+		vec3(0.375, 0.22222222, 0.24),
+		vec3(0.875, 0.55555556, 0.44),
+		vec3(0.0625, 0.88888889, 0.64),
+		vec3(0.5625, 0.03703704, 0.84),
+		vec3(0.3125, 0.37037037, 0.08),
+		vec3(0.8125, 0.7037037, 0.28),
+		vec3(0.1875, 0.14814815, 0.48),
+		vec3(0.6875, 0.48148148, 0.68),
+		vec3(0.4375, 0.81481481, 0.88),
+		vec3(0.9375, 0.25925926, 0.12),
+		vec3(0.03125, 0.59259259, 0.32));
+
+void main() {
+	vec3 fog_cell_size = 1.0 / vec3(scene_params.fog_volume_size);
+
+	ivec3 pos = ivec3(gl_GlobalInvocationID.xyz) + params.corner;
+	if (any(greaterThanEqual(pos, scene_params.fog_volume_size))) {
+		return; //do not compute
+	}
+#ifdef NO_IMAGE_ATOMICS
+	uint lpos = pos.z * scene_params.fog_volume_size.x * scene_params.fog_volume_size.y + pos.y * scene_params.fog_volume_size.x + pos.x;
+#endif
+
+	vec3 posf = vec3(pos);
+
+	vec3 fog_unit_pos = posf * fog_cell_size + fog_cell_size * 0.5; //center of voxels
+	fog_unit_pos.z = pow(fog_unit_pos.z, scene_params.detail_spread);
+
+	vec3 view_pos;
+	view_pos.xy = (fog_unit_pos.xy * 2.0 - 1.0) * mix(scene_params.fog_frustum_size_begin, scene_params.fog_frustum_size_end, vec2(fog_unit_pos.z));
+	view_pos.z = -scene_params.fog_frustum_end * fog_unit_pos.z;
+	view_pos.y = -view_pos.y;
+
+	if (scene_params.use_temporal_reprojection) {
+		vec3 prev_view = (scene_params.to_prev_view * vec4(view_pos, 1.0)).xyz;
+		//undo transform into prev view
+		prev_view.y = -prev_view.y;
+		//z back to unit size
+		prev_view.z /= -scene_params.fog_frustum_end;
+		//xy back to unit size
+		prev_view.xy /= mix(scene_params.fog_frustum_size_begin, scene_params.fog_frustum_size_end, vec2(prev_view.z));
+		prev_view.xy = prev_view.xy * 0.5 + 0.5;
+		//z back to unspread value
+		prev_view.z = pow(prev_view.z, 1.0 / scene_params.detail_spread);
+
+		if (all(greaterThan(prev_view, vec3(0.0))) && all(lessThan(prev_view, vec3(1.0)))) {
+			//reprojectinon fits
+			// Since we can reproject, now we must jitter the current view pos.
+			// This is done here because cells that can't reproject should not jitter.
+
+			fog_unit_pos = posf * fog_cell_size + fog_cell_size * halton_map[scene_params.temporal_frame]; //center of voxels, offset by halton table
+			fog_unit_pos.z = pow(fog_unit_pos.z, scene_params.detail_spread);
+
+			view_pos.xy = (fog_unit_pos.xy * 2.0 - 1.0) * mix(scene_params.fog_frustum_size_begin, scene_params.fog_frustum_size_end, vec2(fog_unit_pos.z));
+			view_pos.z = -scene_params.fog_frustum_end * fog_unit_pos.z;
+			view_pos.y = -view_pos.y;
+		}
+	}
+
+	float density = 0.0;
+	vec3 emission = vec3(0.0);
+	vec3 albedo = vec3(0.0);
+
+	float cell_depth_size = abs(view_pos.z - get_depth_at_pos(fog_cell_size.z, pos.z + 1));
+
+	vec4 world = scene_params.transform * vec4(view_pos, 1.0);
+	world.xyz /= world.w;
+
+	vec3 uvw = fog_unit_pos;
+
+	vec4 local_pos = params.transform * world;
+	local_pos.xyz /= local_pos.w;
+
+	vec3 half_size = params.size / 2.0;
+	float sdf = -1.0;
+	if (params.shape == 0) {
+		// Ellipsoid
+		// https://www.shadertoy.com/view/tdS3DG
+		float k0 = length(local_pos.xyz / half_size);
+		float k1 = length(local_pos.xyz / (half_size * half_size));
+		sdf = k0 * (k0 - 1.0) / k1;
+	} else if (params.shape == 1) {
+		// Cone
+		// https://iquilezles.org/www/articles/distfunctions/distfunctions.htm
+
+		// Compute the cone angle automatically to fit within the volume's size.
+		float inv_height = 1.0 / max(0.001, half_size.y);
+		float radius = 1.0 / max(0.001, (min(half_size.x, half_size.z) * 0.5));
+		float hypotenuse = sqrt(radius * radius + inv_height * inv_height);
+		float rsin = radius / hypotenuse;
+		float rcos = inv_height / hypotenuse;
+		vec2 c = vec2(rsin, rcos);
+
+		float q = length(local_pos.xz);
+		sdf = max(dot(c, vec2(q, local_pos.y - half_size.y)), -half_size.y - local_pos.y);
+	} else if (params.shape == 2) {
+		// Cylinder
+		// https://iquilezles.org/www/articles/distfunctions/distfunctions.htm
+		vec2 d = abs(vec2(length(local_pos.xz), local_pos.y)) - vec2(min(half_size.x, half_size.z), half_size.y);
+		sdf = min(max(d.x, d.y), 0.0) + length(max(d, 0.0));
+	} else if (params.shape == 3) {
+		// Box
+		// https://iquilezles.org/www/articles/distfunctions/distfunctions.htm
+		vec3 q = abs(local_pos.xyz) - half_size;
+		sdf = length(max(q, 0.0)) + min(max(q.x, max(q.y, q.z)), 0.0);
+	}
+
+	float cull_mask = 1.0; //used to cull cells that do not contribute
+	if (params.shape <= 3) {
+#ifndef SDF_USED
+		cull_mask = 1.0 - smoothstep(-0.1, 0.0, sdf);
+#endif
+		uvw = clamp((local_pos.xyz + half_size) / params.size, 0.0, 1.0);
+	}
+
+	if (cull_mask > 0.0) {
+		{
+#CODE : FOG
+		}
+
+#ifdef DENSITY_USED
+		density *= cull_mask;
+		if (abs(density) > 0.001) {
+			int final_density = int(density * DENSITY_SCALE);
+#ifdef NO_IMAGE_ATOMICS
+			atomicAdd(density_only_map[lpos], uint(final_density));
+#else
+			imageAtomicAdd(density_only_map, pos, uint(final_density));
+#endif
+
+#ifdef EMISSION_USED
+			{
+				emission *= clamp(density, 0.0, 1.0);
+				emission = clamp(emission, vec3(0.0), vec3(4.0));
+				// Scale to fit into R11G11B10 with a range of 0-4
+				uvec3 emission_u = uvec3(emission.r * 511.0, emission.g * 511.0, emission.b * 255.0);
+				// R and G have 11 bits each and B has 10. Then pack them into a 32 bit uint
+				uint final_emission = emission_u.r << 21 | emission_u.g << 10 | emission_u.b;
+#ifdef NO_IMAGE_ATOMICS
+				uint prev_emission = atomicAdd(emissive_only_map[lpos], final_emission);
+#else
+				uint prev_emission = imageAtomicAdd(emissive_only_map, pos, final_emission);
+#endif
+
+				// Adding can lead to colors overflowing, so validate
+				uvec3 prev_emission_u = uvec3(prev_emission >> 21, (prev_emission << 11) >> 21, prev_emission % 1024);
+				uint add_emission = final_emission + prev_emission;
+				uvec3 add_emission_u = uvec3(add_emission >> 21, (add_emission << 11) >> 21, add_emission % 1024);
+
+				bvec3 overflowing = lessThan(add_emission_u, prev_emission_u + emission_u);
+
+				if (any(overflowing)) {
+					uvec3 overflow_factor = mix(uvec3(0), uvec3(2047 << 21, 2047 << 10, 1023), overflowing);
+					uint force_max = overflow_factor.r | overflow_factor.g | overflow_factor.b;
+#ifdef NO_IMAGE_ATOMICS
+					atomicOr(emissive_only_map[lpos], force_max);
+#else
+					imageAtomicOr(emissive_only_map, pos, force_max);
+#endif
+				}
+			}
+#endif
+#ifdef ALBEDO_USED
+			{
+				vec3 scattering = albedo * clamp(density, 0.0, 1.0);
+				scattering = clamp(scattering, vec3(0.0), vec3(1.0));
+				uvec3 scattering_u = uvec3(scattering.r * 2047.0, scattering.g * 2047.0, scattering.b * 1023.0);
+				// R and G have 11 bits each and B has 10. Then pack them into a 32 bit uint
+				uint final_scattering = scattering_u.r << 21 | scattering_u.g << 10 | scattering_u.b;
+#ifdef NO_IMAGE_ATOMICS
+				uint prev_scattering = atomicAdd(light_only_map[lpos], final_scattering);
+#else
+				uint prev_scattering = imageAtomicAdd(light_only_map, pos, final_scattering);
+#endif
+
+				// Adding can lead to colors overflowing, so validate
+				uvec3 prev_scattering_u = uvec3(prev_scattering >> 21, (prev_scattering << 11) >> 21, prev_scattering % 1024);
+				uint add_scattering = final_scattering + prev_scattering;
+				uvec3 add_scattering_u = uvec3(add_scattering >> 21, (add_scattering << 11) >> 21, add_scattering % 1024);
+
+				bvec3 overflowing = lessThan(add_scattering_u, prev_scattering_u + scattering_u);
+
+				if (any(overflowing)) {
+					uvec3 overflow_factor = mix(uvec3(0), uvec3(2047 << 21, 2047 << 10, 1023), overflowing);
+					uint force_max = overflow_factor.r | overflow_factor.g | overflow_factor.b;
+#ifdef NO_IMAGE_ATOMICS
+					atomicOr(light_only_map[lpos], force_max);
+#else
+					imageAtomicOr(light_only_map, pos, force_max);
+#endif
+				}
+			}
+#endif // ALBEDO_USED
+		}
+#endif // DENSITY_USED
+	}
+}

servers/rendering/renderer_rd/shaders/environment/volumetric_fog_process.glsl

@@ -0,0 +1,753 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#ifdef USE_VULKAN_MEMORY_MODEL
+#pragma use_vulkan_memory_model
+#endif
+
+#ifdef MODE_DENSITY
+layout(local_size_x = 4, local_size_y = 4, local_size_z = 4) in;
+#else
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+#endif
+
+#include "../cluster_data_inc.glsl"
+#include "../light_data_inc.glsl"
+
+#define M_PI 3.14159265359
+
+#define DENSITY_SCALE 1024.0
+
+layout(set = 0, binding = 1) uniform texture2D shadow_atlas;
+layout(set = 0, binding = 2) uniform texture2D directional_shadow_atlas;
+
+layout(set = 0, binding = 3, std430) restrict readonly buffer OmniLights {
+	LightData data[];
+}
+omni_lights;
+
+layout(set = 0, binding = 4, std430) restrict readonly buffer SpotLights {
+	LightData data[];
+}
+spot_lights;
+
+layout(set = 0, binding = 5, std140) uniform DirectionalLights {
+	DirectionalLightData data[MAX_DIRECTIONAL_LIGHT_DATA_STRUCTS];
+}
+directional_lights;
+
+layout(set = 0, binding = 6, std430) buffer restrict readonly ClusterBuffer {
+	uint data[];
+}
+cluster_buffer;
+
+layout(set = 0, binding = 7) uniform sampler linear_sampler;
+
+#ifdef MODE_DENSITY
+layout(rgba16f, set = 0, binding = 8) uniform restrict writeonly image3D density_map;
+#endif
+
+#ifdef MODE_FOG
+layout(rgba16f, set = 0, binding = 8) uniform restrict readonly image3D density_map;
+layout(rgba16f, set = 0, binding = 9) uniform restrict writeonly image3D fog_map;
+#endif
+
+#ifdef MODE_COPY
+layout(rgba16f, set = 0, binding = 8) uniform restrict readonly image3D source_map;
+layout(rgba16f, set = 0, binding = 9) uniform restrict writeonly image3D dest_map;
+#endif
+
+#ifdef MODE_FILTER
+layout(rgba16f, set = 0, binding = 8) uniform restrict readonly image3D source_map;
+layout(rgba16f, set = 0, binding = 9) uniform restrict writeonly image3D dest_map;
+#endif
+
+layout(set = 0, binding = 10) uniform sampler shadow_sampler;
+
+#define MAX_VOXEL_GI_INSTANCES 8
+
+struct VoxelGIData {
+	mat4 xform; // 64 - 64
+
+	vec3 bounds; // 12 - 76
+	float dynamic_range; // 4 - 80
+
+	float bias; // 4 - 84
+	float normal_bias; // 4 - 88
+	bool blend_ambient; // 4 - 92
+	uint mipmaps; // 4 - 96
+
+	vec3 pad; // 12 - 108
+	float exposure_normalization; // 4 - 112
+};
+
+layout(set = 0, binding = 11, std140) uniform VoxelGIs {
+	VoxelGIData data[MAX_VOXEL_GI_INSTANCES];
+}
+voxel_gi_instances;
+
+layout(set = 0, binding = 12) uniform texture3D voxel_gi_textures[MAX_VOXEL_GI_INSTANCES];
+
+layout(set = 0, binding = 13) uniform sampler linear_sampler_with_mipmaps;
+
+#ifdef ENABLE_SDFGI
+
+// SDFGI Integration on set 1
+#define SDFGI_MAX_CASCADES 8
+
+struct SDFVoxelGICascadeData {
+	vec3 position;
+	float to_probe;
+	ivec3 probe_world_offset;
+	float to_cell; // 1/bounds * grid_size
+	vec3 pad;
+	float exposure_normalization;
+};
+
+layout(set = 1, binding = 0, std140) uniform SDFGI {
+	vec3 grid_size;
+	uint max_cascades;
+
+	bool use_occlusion;
+	int probe_axis_size;
+	float probe_to_uvw;
+	float normal_bias;
+
+	vec3 lightprobe_tex_pixel_size;
+	float energy;
+
+	vec3 lightprobe_uv_offset;
+	float y_mult;
+
+	vec3 occlusion_clamp;
+	uint pad3;
+
+	vec3 occlusion_renormalize;
+	uint pad4;
+
+	vec3 cascade_probe_size;
+	uint pad5;
+
+	SDFVoxelGICascadeData cascades[SDFGI_MAX_CASCADES];
+}
+sdfgi;
+
+layout(set = 1, binding = 1) uniform texture2DArray sdfgi_ambient_texture;
+
+layout(set = 1, binding = 2) uniform texture3D sdfgi_occlusion_texture;
+
+#endif //SDFGI
+
+layout(set = 0, binding = 14, std140) uniform Params {
+	vec2 fog_frustum_size_begin;
+	vec2 fog_frustum_size_end;
+
+	float fog_frustum_end;
+	float ambient_inject;
+	float z_far;
+	int filter_axis;
+
+	vec3 ambient_color;
+	float sky_contribution;
+
+	ivec3 fog_volume_size;
+	uint directional_light_count;
+
+	vec3 base_emission;
+	float base_density;
+
+	vec3 base_scattering;
+	float phase_g;
+
+	float detail_spread;
+	float gi_inject;
+	uint max_voxel_gi_instances;
+	uint cluster_type_size;
+
+	vec2 screen_size;
+	uint cluster_shift;
+	uint cluster_width;
+
+	uint max_cluster_element_count_div_32;
+	bool use_temporal_reprojection;
+	uint temporal_frame;
+	float temporal_blend;
+
+	mat3x4 cam_rotation;
+	mat4 to_prev_view;
+
+	mat3 radiance_inverse_xform;
+}
+params;
+#ifndef MODE_COPY
+layout(set = 0, binding = 15) uniform texture3D prev_density_texture;
+
+#ifdef NO_IMAGE_ATOMICS
+layout(set = 0, binding = 16) buffer density_only_map_buffer {
+	uint density_only_map[];
+};
+layout(set = 0, binding = 17) buffer light_only_map_buffer {
+	uint light_only_map[];
+};
+layout(set = 0, binding = 18) buffer emissive_only_map_buffer {
+	uint emissive_only_map[];
+};
+#else
+layout(r32ui, set = 0, binding = 16) uniform uimage3D density_only_map;
+layout(r32ui, set = 0, binding = 17) uniform uimage3D light_only_map;
+layout(r32ui, set = 0, binding = 18) uniform uimage3D emissive_only_map;
+#endif
+
+#ifdef USE_RADIANCE_CUBEMAP_ARRAY
+layout(set = 0, binding = 19) uniform textureCubeArray sky_texture;
+#else
+layout(set = 0, binding = 19) uniform textureCube sky_texture;
+#endif
+#endif // MODE_COPY
+
+float get_depth_at_pos(float cell_depth_size, int z) {
+	float d = float(z) * cell_depth_size + cell_depth_size * 0.5; //center of voxels
+	d = pow(d, params.detail_spread);
+	return params.fog_frustum_end * d;
+}
+
+vec3 hash3f(uvec3 x) {
+	x = ((x >> 16) ^ x) * 0x45d9f3b;
+	x = ((x >> 16) ^ x) * 0x45d9f3b;
+	x = (x >> 16) ^ x;
+	return vec3(x & 0xFFFFF) / vec3(float(0xFFFFF));
+}
+
+float get_omni_attenuation(float dist, float inv_range, float decay) {
+	float nd = dist * inv_range;
+	nd *= nd;
+	nd *= nd; // nd^4
+	nd = max(1.0 - nd, 0.0);
+	nd *= nd; // nd^2
+	return nd * pow(max(dist, 0.0001), -decay);
+}
+
+void cluster_get_item_range(uint p_offset, out uint item_min, out uint item_max, out uint item_from, out uint item_to) {
+	uint item_min_max = cluster_buffer.data[p_offset];
+	item_min = item_min_max & 0xFFFF;
+	item_max = item_min_max >> 16;
+
+	item_from = item_min >> 5;
+	item_to = (item_max == 0) ? 0 : ((item_max - 1) >> 5) + 1; //side effect of how it is stored, as item_max 0 means no elements
+}
+
+uint cluster_get_range_clip_mask(uint i, uint z_min, uint z_max) {
+	int local_min = clamp(int(z_min) - int(i) * 32, 0, 31);
+	int mask_width = min(int(z_max) - int(z_min), 32 - local_min);
+	return bitfieldInsert(uint(0), uint(0xFFFFFFFF), local_min, mask_width);
+}
+
+float henyey_greenstein(float cos_theta, float g) {
+	const float k = 0.0795774715459; // 1 / (4 * PI)
+	return k * (1.0 - g * g) / (pow(1.0 + g * g - 2.0 * g * cos_theta, 1.5));
+}
+
+#define TEMPORAL_FRAMES 16
+
+const vec3 halton_map[TEMPORAL_FRAMES] = vec3[](
+		vec3(0.5, 0.33333333, 0.2),
+		vec3(0.25, 0.66666667, 0.4),
+		vec3(0.75, 0.11111111, 0.6),
+		vec3(0.125, 0.44444444, 0.8),
+		vec3(0.625, 0.77777778, 0.04),
+		vec3(0.375, 0.22222222, 0.24),
+		vec3(0.875, 0.55555556, 0.44),
+		vec3(0.0625, 0.88888889, 0.64),
+		vec3(0.5625, 0.03703704, 0.84),
+		vec3(0.3125, 0.37037037, 0.08),
+		vec3(0.8125, 0.7037037, 0.28),
+		vec3(0.1875, 0.14814815, 0.48),
+		vec3(0.6875, 0.48148148, 0.68),
+		vec3(0.4375, 0.81481481, 0.88),
+		vec3(0.9375, 0.25925926, 0.12),
+		vec3(0.03125, 0.59259259, 0.32));
+
+// Higher values will make light in volumetric fog fade out sooner when it's occluded by shadow.
+const float INV_FOG_FADE = 10.0;
+
+void main() {
+	vec3 fog_cell_size = 1.0 / vec3(params.fog_volume_size);
+
+#ifdef MODE_DENSITY
+
+	ivec3 pos = ivec3(gl_GlobalInvocationID.xyz);
+	if (any(greaterThanEqual(pos, params.fog_volume_size))) {
+		return; //do not compute
+	}
+#ifdef NO_IMAGE_ATOMICS
+	uint lpos = pos.z * params.fog_volume_size.x * params.fog_volume_size.y + pos.y * params.fog_volume_size.x + pos.x;
+#endif
+
+	vec3 posf = vec3(pos);
+
+	//posf += mix(vec3(0.0),vec3(1.0),0.3) * hash3f(uvec3(pos)) * 2.0 - 1.0;
+
+	vec3 fog_unit_pos = posf * fog_cell_size + fog_cell_size * 0.5; //center of voxels
+
+	uvec2 screen_pos = uvec2(fog_unit_pos.xy * params.screen_size);
+	uvec2 cluster_pos = screen_pos >> params.cluster_shift;
+	uint cluster_offset = (params.cluster_width * cluster_pos.y + cluster_pos.x) * (params.max_cluster_element_count_div_32 + 32);
+	//positions in screen are too spread apart, no hopes for optimizing with subgroups
+
+	fog_unit_pos.z = pow(fog_unit_pos.z, params.detail_spread);
+
+	vec3 view_pos;
+	view_pos.xy = (fog_unit_pos.xy * 2.0 - 1.0) * mix(params.fog_frustum_size_begin, params.fog_frustum_size_end, vec2(fog_unit_pos.z));
+	view_pos.z = -params.fog_frustum_end * fog_unit_pos.z;
+	view_pos.y = -view_pos.y;
+
+	vec4 reprojected_density = vec4(0.0);
+	float reproject_amount = 0.0;
+
+	if (params.use_temporal_reprojection) {
+		vec3 prev_view = (params.to_prev_view * vec4(view_pos, 1.0)).xyz;
+		//undo transform into prev view
+		prev_view.y = -prev_view.y;
+		//z back to unit size
+		prev_view.z /= -params.fog_frustum_end;
+		//xy back to unit size
+		prev_view.xy /= mix(params.fog_frustum_size_begin, params.fog_frustum_size_end, vec2(prev_view.z));
+		prev_view.xy = prev_view.xy * 0.5 + 0.5;
+		//z back to unspread value
+		prev_view.z = pow(prev_view.z, 1.0 / params.detail_spread);
+
+		if (all(greaterThan(prev_view, vec3(0.0))) && all(lessThan(prev_view, vec3(1.0)))) {
+			//reprojectinon fits
+
+			reprojected_density = textureLod(sampler3D(prev_density_texture, linear_sampler), prev_view, 0.0);
+			reproject_amount = params.temporal_blend;
+
+			// Since we can reproject, now we must jitter the current view pos.
+			// This is done here because cells that can't reproject should not jitter.
+
+			fog_unit_pos = posf * fog_cell_size + fog_cell_size * halton_map[params.temporal_frame]; //center of voxels, offset by halton table
+
+			screen_pos = uvec2(fog_unit_pos.xy * params.screen_size);
+			cluster_pos = screen_pos >> params.cluster_shift;
+			cluster_offset = (params.cluster_width * cluster_pos.y + cluster_pos.x) * (params.max_cluster_element_count_div_32 + 32);
+			//positions in screen are too spread apart, no hopes for optimizing with subgroups
+
+			fog_unit_pos.z = pow(fog_unit_pos.z, params.detail_spread);
+
+			view_pos.xy = (fog_unit_pos.xy * 2.0 - 1.0) * mix(params.fog_frustum_size_begin, params.fog_frustum_size_end, vec2(fog_unit_pos.z));
+			view_pos.z = -params.fog_frustum_end * fog_unit_pos.z;
+			view_pos.y = -view_pos.y;
+		}
+	}
+
+	uint cluster_z = uint(clamp((abs(view_pos.z) / params.z_far) * 32.0, 0.0, 31.0));
+
+	vec3 total_light = vec3(0.0);
+
+	float total_density = params.base_density;
+#ifdef NO_IMAGE_ATOMICS
+	uint local_density = density_only_map[lpos];
+#else
+	uint local_density = imageLoad(density_only_map, pos).x;
+#endif
+
+	total_density += float(int(local_density)) / DENSITY_SCALE;
+	total_density = max(0.0, total_density);
+
+#ifdef NO_IMAGE_ATOMICS
+	uint scattering_u = light_only_map[lpos];
+#else
+	uint scattering_u = imageLoad(light_only_map, pos).x;
+#endif
+	vec3 scattering = vec3(scattering_u >> 21, (scattering_u << 11) >> 21, scattering_u % 1024) / vec3(2047.0, 2047.0, 1023.0);
+	scattering += params.base_scattering * params.base_density;
+
+#ifdef NO_IMAGE_ATOMICS
+	uint emission_u = emissive_only_map[lpos];
+#else
+	uint emission_u = imageLoad(emissive_only_map, pos).x;
+#endif
+	vec3 emission = vec3(emission_u >> 21, (emission_u << 11) >> 21, emission_u % 1024) / vec3(511.0, 511.0, 255.0);
+	emission += params.base_emission * params.base_density;
+
+	float cell_depth_size = abs(view_pos.z - get_depth_at_pos(fog_cell_size.z, pos.z + 1));
+	//compute directional lights
+
+	if (total_density > 0.00005) {
+		for (uint i = 0; i < params.directional_light_count; i++) {
+			if (directional_lights.data[i].volumetric_fog_energy > 0.001) {
+				vec3 shadow_attenuation = vec3(1.0);
+
+				if (directional_lights.data[i].shadow_opacity > 0.001) {
+					float depth_z = -view_pos.z;
+
+					vec4 pssm_coord;
+					vec3 light_dir = directional_lights.data[i].direction;
+					vec4 v = vec4(view_pos, 1.0);
+					float z_range;
+
+					if (depth_z < directional_lights.data[i].shadow_split_offsets.x) {
+						pssm_coord = (directional_lights.data[i].shadow_matrix1 * v);
+						pssm_coord /= pssm_coord.w;
+						z_range = directional_lights.data[i].shadow_z_range.x;
+
+					} else if (depth_z < directional_lights.data[i].shadow_split_offsets.y) {
+						pssm_coord = (directional_lights.data[i].shadow_matrix2 * v);
+						pssm_coord /= pssm_coord.w;
+						z_range = directional_lights.data[i].shadow_z_range.y;
+
+					} else if (depth_z < directional_lights.data[i].shadow_split_offsets.z) {
+						pssm_coord = (directional_lights.data[i].shadow_matrix3 * v);
+						pssm_coord /= pssm_coord.w;
+						z_range = directional_lights.data[i].shadow_z_range.z;
+
+					} else {
+						pssm_coord = (directional_lights.data[i].shadow_matrix4 * v);
+						pssm_coord /= pssm_coord.w;
+						z_range = directional_lights.data[i].shadow_z_range.w;
+					}
+
+					float depth = texture(sampler2D(directional_shadow_atlas, linear_sampler), pssm_coord.xy).r;
+					float shadow = exp(min(0.0, (pssm_coord.z - depth)) * z_range * INV_FOG_FADE);
+
+					shadow = mix(shadow, 1.0, smoothstep(directional_lights.data[i].fade_from, directional_lights.data[i].fade_to, view_pos.z)); //done with negative values for performance
+
+					shadow_attenuation = mix(vec3(1.0 - directional_lights.data[i].shadow_opacity), vec3(1.0), shadow);
+				}
+
+				total_light += shadow_attenuation * directional_lights.data[i].color * directional_lights.data[i].energy * henyey_greenstein(dot(normalize(view_pos), normalize(directional_lights.data[i].direction)), params.phase_g) * directional_lights.data[i].volumetric_fog_energy;
+			}
+		}
+
+		// Compute light from sky
+		if (params.ambient_inject > 0.0) {
+			vec3 isotropic = vec3(0.0);
+			vec3 anisotropic = vec3(0.0);
+			if (params.sky_contribution > 0.0) {
+				float mip_bias = 2.0 + total_density * (MAX_SKY_LOD - 2.0); // Not physically based, but looks nice
+				vec3 scatter_direction = (params.radiance_inverse_xform * normalize(view_pos)) * sign(params.phase_g);
+#ifdef USE_RADIANCE_CUBEMAP_ARRAY
+				isotropic = texture(samplerCubeArray(sky_texture, linear_sampler_with_mipmaps), vec4(0.0, 1.0, 0.0, mip_bias)).rgb;
+				anisotropic = texture(samplerCubeArray(sky_texture, linear_sampler_with_mipmaps), vec4(scatter_direction, mip_bias)).rgb;
+#else
+				isotropic = textureLod(samplerCube(sky_texture, linear_sampler_with_mipmaps), vec3(0.0, 1.0, 0.0), mip_bias).rgb;
+				anisotropic = textureLod(samplerCube(sky_texture, linear_sampler_with_mipmaps), vec3(scatter_direction), mip_bias).rgb;
+#endif //USE_RADIANCE_CUBEMAP_ARRAY
+			}
+
+			total_light += mix(params.ambient_color, mix(isotropic, anisotropic, abs(params.phase_g)), params.sky_contribution) * params.ambient_inject;
+		}
+
+		//compute lights from cluster
+
+		{ //omni lights
+
+			uint cluster_omni_offset = cluster_offset;
+
+			uint item_min;
+			uint item_max;
+			uint item_from;
+			uint item_to;
+
+			cluster_get_item_range(cluster_omni_offset + params.max_cluster_element_count_div_32 + cluster_z, item_min, item_max, item_from, item_to);
+
+			for (uint i = item_from; i < item_to; i++) {
+				uint mask = cluster_buffer.data[cluster_omni_offset + i];
+				mask &= cluster_get_range_clip_mask(i, item_min, item_max);
+				uint merged_mask = mask;
+
+				while (merged_mask != 0) {
+					uint bit = findMSB(merged_mask);
+					merged_mask &= ~(1 << bit);
+
+					uint light_index = 32 * i + bit;
+
+					//if (!bool(omni_omni_lights.data[light_index].mask & draw_call.layer_mask)) {
+					//	continue; //not masked
+					//}
+
+					vec3 light_pos = omni_lights.data[light_index].position;
+					float d = distance(omni_lights.data[light_index].position, view_pos);
+					float shadow_attenuation = 1.0;
+
+					if (omni_lights.data[light_index].volumetric_fog_energy > 0.001 && d * omni_lights.data[light_index].inv_radius < 1.0) {
+						float attenuation = get_omni_attenuation(d, omni_lights.data[light_index].inv_radius, omni_lights.data[light_index].attenuation);
+
+						vec3 light = omni_lights.data[light_index].color;
+
+						if (omni_lights.data[light_index].shadow_opacity > 0.001) {
+							//has shadow
+							vec4 uv_rect = omni_lights.data[light_index].atlas_rect;
+							vec2 flip_offset = omni_lights.data[light_index].direction.xy;
+
+							vec3 local_vert = (omni_lights.data[light_index].shadow_matrix * vec4(view_pos, 1.0)).xyz;
+
+							float shadow_len = length(local_vert); //need to remember shadow len from here
+							vec3 shadow_sample = normalize(local_vert);
+
+							if (shadow_sample.z >= 0.0) {
+								uv_rect.xy += flip_offset;
+							}
+
+							shadow_sample.z = 1.0 + abs(shadow_sample.z);
+							vec3 pos = vec3(shadow_sample.xy / shadow_sample.z, shadow_len - omni_lights.data[light_index].shadow_bias);
+							pos.z *= omni_lights.data[light_index].inv_radius;
+							pos.z = 1.0 - pos.z;
+
+							pos.xy = pos.xy * 0.5 + 0.5;
+							pos.xy = uv_rect.xy + pos.xy * uv_rect.zw;
+
+							float depth = texture(sampler2D(shadow_atlas, linear_sampler), pos.xy).r;
+
+							shadow_attenuation = mix(1.0 - omni_lights.data[light_index].shadow_opacity, 1.0, exp(min(0.0, (pos.z - depth)) / omni_lights.data[light_index].inv_radius * INV_FOG_FADE));
+						}
+						total_light += light * attenuation * shadow_attenuation * henyey_greenstein(dot(normalize(light_pos - view_pos), normalize(view_pos)), params.phase_g) * omni_lights.data[light_index].volumetric_fog_energy;
+					}
+				}
+			}
+		}
+
+		{ //spot lights
+
+			uint cluster_spot_offset = cluster_offset + params.cluster_type_size;
+
+			uint item_min;
+			uint item_max;
+			uint item_from;
+			uint item_to;
+
+			cluster_get_item_range(cluster_spot_offset + params.max_cluster_element_count_div_32 + cluster_z, item_min, item_max, item_from, item_to);
+
+			for (uint i = item_from; i < item_to; i++) {
+				uint mask = cluster_buffer.data[cluster_spot_offset + i];
+				mask &= cluster_get_range_clip_mask(i, item_min, item_max);
+				uint merged_mask = mask;
+
+				while (merged_mask != 0) {
+					uint bit = findMSB(merged_mask);
+					merged_mask &= ~(1 << bit);
+
+					//if (!bool(omni_lights.data[light_index].mask & draw_call.layer_mask)) {
+					//	continue; //not masked
+					//}
+
+					uint light_index = 32 * i + bit;
+
+					vec3 light_pos = spot_lights.data[light_index].position;
+					vec3 light_rel_vec = spot_lights.data[light_index].position - view_pos;
+					float d = length(light_rel_vec);
+					float shadow_attenuation = 1.0;
+
+					if (spot_lights.data[light_index].volumetric_fog_energy > 0.001 && d * spot_lights.data[light_index].inv_radius < 1.0) {
+						float attenuation = get_omni_attenuation(d, spot_lights.data[light_index].inv_radius, spot_lights.data[light_index].attenuation);
+
+						vec3 spot_dir = spot_lights.data[light_index].direction;
+						float cone_angle = spot_lights.data[light_index].cone_angle;
+						float scos = max(dot(-normalize(light_rel_vec), spot_dir), cone_angle);
+						float spot_rim = max(0.0001, (1.0 - scos) / (1.0 - cone_angle));
+						attenuation *= 1.0 - pow(spot_rim, spot_lights.data[light_index].cone_attenuation);
+
+						vec3 light = spot_lights.data[light_index].color;
+
+						if (spot_lights.data[light_index].shadow_opacity > 0.001) {
+							//has shadow
+							vec4 uv_rect = spot_lights.data[light_index].atlas_rect;
+
+							vec4 v = vec4(view_pos, 1.0);
+
+							vec4 splane = (spot_lights.data[light_index].shadow_matrix * v);
+							splane.z -= spot_lights.data[light_index].shadow_bias / (d * spot_lights.data[light_index].inv_radius);
+							splane /= splane.w;
+
+							vec3 pos = vec3(splane.xy * spot_lights.data[light_index].atlas_rect.zw + spot_lights.data[light_index].atlas_rect.xy, splane.z);
+
+							float depth = texture(sampler2D(shadow_atlas, linear_sampler), pos.xy).r;
+
+							shadow_attenuation = mix(1.0 - spot_lights.data[light_index].shadow_opacity, 1.0, exp(min(0.0, (pos.z - depth)) / spot_lights.data[light_index].inv_radius * INV_FOG_FADE));
+						}
+						total_light += light * attenuation * shadow_attenuation * henyey_greenstein(dot(normalize(light_rel_vec), normalize(view_pos)), params.phase_g) * spot_lights.data[light_index].volumetric_fog_energy;
+					}
+				}
+			}
+		}
+
+		vec3 world_pos = mat3(params.cam_rotation) * view_pos;
+
+		for (uint i = 0; i < params.max_voxel_gi_instances; i++) {
+			vec3 position = (voxel_gi_instances.data[i].xform * vec4(world_pos, 1.0)).xyz;
+
+			//this causes corrupted pixels, i have no idea why..
+			if (all(bvec2(all(greaterThanEqual(position, vec3(0.0))), all(lessThan(position, voxel_gi_instances.data[i].bounds))))) {
+				position /= voxel_gi_instances.data[i].bounds;
+
+				vec4 light = vec4(0.0);
+				for (uint j = 0; j < voxel_gi_instances.data[i].mipmaps; j++) {
+					vec4 slight = textureLod(sampler3D(voxel_gi_textures[i], linear_sampler_with_mipmaps), position, float(j));
+					float a = (1.0 - light.a);
+					light += a * slight;
+				}
+
+				light.rgb *= voxel_gi_instances.data[i].dynamic_range * params.gi_inject * voxel_gi_instances.data[i].exposure_normalization;
+
+				total_light += light.rgb;
+			}
+		}
+
+		//sdfgi
+#ifdef ENABLE_SDFGI
+
+		{
+			float blend = -1.0;
+			vec3 ambient_total = vec3(0.0);
+
+			for (uint i = 0; i < sdfgi.max_cascades; i++) {
+				vec3 cascade_pos = (world_pos - sdfgi.cascades[i].position) * sdfgi.cascades[i].to_probe;
+
+				if (any(lessThan(cascade_pos, vec3(0.0))) || any(greaterThanEqual(cascade_pos, sdfgi.cascade_probe_size))) {
+					continue; //skip cascade
+				}
+
+				vec3 base_pos = floor(cascade_pos);
+				ivec3 probe_base_pos = ivec3(base_pos);
+
+				vec4 ambient_accum = vec4(0.0);
+
+				ivec3 tex_pos = ivec3(probe_base_pos.xy, int(i));
+				tex_pos.x += probe_base_pos.z * sdfgi.probe_axis_size;
+
+				for (uint j = 0; j < 8; j++) {
+					ivec3 offset = (ivec3(j) >> ivec3(0, 1, 2)) & ivec3(1, 1, 1);
+					ivec3 probe_posi = probe_base_pos;
+					probe_posi += offset;
+
+					// Compute weight
+
+					vec3 probe_pos = vec3(probe_posi);
+					vec3 probe_to_pos = cascade_pos - probe_pos;
+
+					vec3 trilinear = vec3(1.0) - abs(probe_to_pos);
+					float weight = trilinear.x * trilinear.y * trilinear.z;
+
+					// Compute lightprobe occlusion
+
+					if (sdfgi.use_occlusion) {
+						ivec3 occ_indexv = abs((sdfgi.cascades[i].probe_world_offset + probe_posi) & ivec3(1, 1, 1)) * ivec3(1, 2, 4);
+						vec4 occ_mask = mix(vec4(0.0), vec4(1.0), equal(ivec4(occ_indexv.x | occ_indexv.y), ivec4(0, 1, 2, 3)));
+
+						vec3 occ_pos = clamp(cascade_pos, probe_pos - sdfgi.occlusion_clamp, probe_pos + sdfgi.occlusion_clamp) * sdfgi.probe_to_uvw;
+						occ_pos.z += float(i);
+						if (occ_indexv.z != 0) { //z bit is on, means index is >=4, so make it switch to the other half of textures
+							occ_pos.x += 1.0;
+						}
+
+						occ_pos *= sdfgi.occlusion_renormalize;
+						float occlusion = dot(textureLod(sampler3D(sdfgi_occlusion_texture, linear_sampler), occ_pos, 0.0), occ_mask);
+
+						weight *= max(occlusion, 0.01);
+					}
+
+					// Compute ambient texture position
+
+					ivec3 uvw = tex_pos;
+					uvw.xy += offset.xy;
+					uvw.x += offset.z * sdfgi.probe_axis_size;
+
+					vec3 ambient = texelFetch(sampler2DArray(sdfgi_ambient_texture, linear_sampler), uvw, 0).rgb;
+
+					ambient_accum.rgb += ambient * weight * sdfgi.cascades[i].exposure_normalization;
+					ambient_accum.a += weight;
+				}
+
+				if (ambient_accum.a > 0) {
+					ambient_accum.rgb /= ambient_accum.a;
+				}
+				ambient_total = ambient_accum.rgb;
+				break;
+			}
+
+			total_light += ambient_total * params.gi_inject;
+		}
+
+#endif
+	}
+
+	vec4 final_density = vec4(total_light * scattering + emission, total_density);
+
+	final_density = mix(final_density, reprojected_density, reproject_amount);
+
+	imageStore(density_map, pos, final_density);
+#ifdef NO_IMAGE_ATOMICS
+	density_only_map[lpos] = 0;
+	light_only_map[lpos] = 0;
+	emissive_only_map[lpos] = 0;
+#else
+	imageStore(density_only_map, pos, uvec4(0));
+	imageStore(light_only_map, pos, uvec4(0));
+	imageStore(emissive_only_map, pos, uvec4(0));
+#endif
+#endif
+
+#ifdef MODE_FOG
+
+	ivec3 pos = ivec3(gl_GlobalInvocationID.xy, 0);
+
+	if (any(greaterThanEqual(pos, params.fog_volume_size))) {
+		return; //do not compute
+	}
+
+	vec4 fog_accum = vec4(0.0, 0.0, 0.0, 1.0);
+	float prev_z = 0.0;
+
+	for (int i = 0; i < params.fog_volume_size.z; i++) {
+		//compute fog position
+		ivec3 fog_pos = pos + ivec3(0, 0, i);
+		//get fog value
+		vec4 fog = imageLoad(density_map, fog_pos);
+
+		//get depth at cell pos
+		float z = get_depth_at_pos(fog_cell_size.z, i);
+		//get distance from previous pos
+		float d = abs(prev_z - z);
+		//compute transmittance using beer's law
+		float transmittance = exp(-d * fog.a);
+
+		fog_accum.rgb += ((fog.rgb - fog.rgb * transmittance) / max(fog.a, 0.00001)) * fog_accum.a;
+		fog_accum.a *= transmittance;
+
+		prev_z = z;
+
+		imageStore(fog_map, fog_pos, vec4(fog_accum.rgb, 1.0 - fog_accum.a));
+	}
+
+#endif
+
+#ifdef MODE_FILTER
+
+	ivec3 pos = ivec3(gl_GlobalInvocationID.xyz);
+
+	const float gauss[7] = float[](0.071303, 0.131514, 0.189879, 0.214607, 0.189879, 0.131514, 0.071303);
+
+	const ivec3 filter_dir[3] = ivec3[](ivec3(1, 0, 0), ivec3(0, 1, 0), ivec3(0, 0, 1));
+	ivec3 offset = filter_dir[params.filter_axis];
+
+	vec4 accum = vec4(0.0);
+	for (int i = -3; i <= 3; i++) {
+		accum += imageLoad(source_map, clamp(pos + offset * i, ivec3(0), params.fog_volume_size - ivec3(1))) * gauss[i + 3];
+	}
+
+	imageStore(dest_map, pos, accum);
+
+#endif
+#ifdef MODE_COPY
+	ivec3 pos = ivec3(gl_GlobalInvocationID.xyz);
+	if (any(greaterThanEqual(pos, params.fog_volume_size))) {
+		return; //do not compute
+	}
+
+	imageStore(dest_map, pos, imageLoad(source_map, pos));
+
+#endif
+}

servers/rendering/renderer_rd/shaders/environment/voxel_gi.glsl

@@ -0,0 +1,620 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#ifdef MODE_DYNAMIC
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+#else
+layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
+#endif
+
+#ifndef MODE_DYNAMIC
+
+#define NO_CHILDREN 0xFFFFFFFF
+
+struct CellChildren {
+	uint children[8];
+};
+
+layout(set = 0, binding = 1, std430) buffer CellChildrenBuffer {
+	CellChildren data[];
+}
+cell_children;
+
+struct CellData {
+	uint position; // xyz 10 bits
+	uint albedo; //rgb albedo
+	uint emission; //rgb normalized with e as multiplier
+	uint normal; //RGB normal encoded
+};
+
+layout(set = 0, binding = 2, std430) buffer CellDataBuffer {
+	CellData data[];
+}
+cell_data;
+
+#endif // MODE DYNAMIC
+
+#define LIGHT_TYPE_DIRECTIONAL 0
+#define LIGHT_TYPE_OMNI 1
+#define LIGHT_TYPE_SPOT 2
+
+#if defined(MODE_COMPUTE_LIGHT) || defined(MODE_DYNAMIC_LIGHTING)
+
+struct Light {
+	uint type;
+	float energy;
+	float radius;
+	float attenuation;
+
+	vec3 color;
+	float cos_spot_angle;
+
+	vec3 position;
+	float inv_spot_attenuation;
+
+	vec3 direction;
+	bool has_shadow;
+};
+
+layout(set = 0, binding = 3, std140) uniform Lights {
+	Light data[MAX_LIGHTS];
+}
+lights;
+
+#endif // MODE COMPUTE LIGHT
+
+#ifdef MODE_SECOND_BOUNCE
+
+layout(set = 0, binding = 5) uniform texture3D color_texture;
+
+#endif // MODE_SECOND_BOUNCE
+
+#ifndef MODE_DYNAMIC
+
+layout(push_constant, std430) uniform Params {
+	ivec3 limits;
+	uint stack_size;
+
+	float emission_scale;
+	float propagation;
+	float dynamic_range;
+
+	uint light_count;
+	uint cell_offset;
+	uint cell_count;
+	float aniso_strength;
+	float cell_size;
+}
+params;
+
+layout(set = 0, binding = 4, std430) buffer Outputs {
+	vec4 data[];
+}
+outputs;
+
+#endif // MODE DYNAMIC
+
+layout(set = 0, binding = 9) uniform texture3D texture_sdf;
+layout(set = 0, binding = 10) uniform sampler texture_sampler;
+
+#ifdef MODE_WRITE_TEXTURE
+
+layout(rgba8, set = 0, binding = 5) uniform restrict writeonly image3D color_tex;
+
+#endif
+
+#ifdef MODE_DYNAMIC
+
+layout(push_constant, std430) uniform Params {
+	ivec3 limits;
+	uint light_count; //when not lighting
+	ivec3 x_dir;
+	float z_base;
+	ivec3 y_dir;
+	float z_sign;
+	ivec3 z_dir;
+	float pos_multiplier;
+	ivec2 rect_pos;
+	ivec2 rect_size;
+	ivec2 prev_rect_ofs;
+	ivec2 prev_rect_size;
+	bool flip_x;
+	bool flip_y;
+	float dynamic_range;
+	bool on_mipmap;
+	float propagation;
+	float cell_size;
+	float pad[2];
+}
+params;
+
+#ifdef MODE_DYNAMIC_LIGHTING
+
+layout(rgba8, set = 0, binding = 5) uniform restrict readonly image2D source_albedo;
+layout(rgba8, set = 0, binding = 6) uniform restrict readonly image2D source_normal;
+layout(rgba8, set = 0, binding = 7) uniform restrict readonly image2D source_orm;
+//layout (set=0,binding=8) uniform texture2D source_depth;
+layout(rgba16f, set = 0, binding = 11) uniform restrict image2D emission;
+layout(r32f, set = 0, binding = 12) uniform restrict image2D depth;
+
+#endif
+
+#ifdef MODE_DYNAMIC_SHRINK
+
+layout(rgba16f, set = 0, binding = 5) uniform restrict readonly image2D source_light;
+layout(r32f, set = 0, binding = 6) uniform restrict readonly image2D source_depth;
+
+#ifdef MODE_DYNAMIC_SHRINK_WRITE
+
+layout(rgba16f, set = 0, binding = 7) uniform restrict writeonly image2D light;
+layout(r32f, set = 0, binding = 8) uniform restrict writeonly image2D depth;
+
+#endif // MODE_DYNAMIC_SHRINK_WRITE
+
+#ifdef MODE_DYNAMIC_SHRINK_PLOT
+
+layout(rgba8, set = 0, binding = 11) uniform restrict image3D color_texture;
+
+#endif //MODE_DYNAMIC_SHRINK_PLOT
+
+#endif // MODE_DYNAMIC_SHRINK
+
+//layout (rgba8,set=0,binding=5) uniform restrict writeonly image3D color_tex;
+
+#endif // MODE DYNAMIC
+
+#if defined(MODE_COMPUTE_LIGHT) || defined(MODE_DYNAMIC_LIGHTING)
+
+float raymarch(float distance, float distance_adv, vec3 from, vec3 direction) {
+	vec3 cell_size = 1.0 / vec3(params.limits);
+	float occlusion = 1.0;
+	while (distance > 0.5) { //use this to avoid precision errors
+		float advance = texture(sampler3D(texture_sdf, texture_sampler), from * cell_size).r * 255.0 - 1.0;
+		if (advance < 0.0) {
+			occlusion = 0.0;
+			break;
+		}
+
+		occlusion = min(advance, occlusion);
+
+		advance = max(distance_adv, advance - mod(advance, distance_adv)); //should always advance in multiples of distance_adv
+
+		from += direction * advance;
+		distance -= advance;
+	}
+
+	return occlusion; //max(0.0,distance);
+}
+
+float get_omni_attenuation(float distance, float inv_range, float decay) {
+	float nd = distance * inv_range;
+	nd *= nd;
+	nd *= nd; // nd^4
+	nd = max(1.0 - nd, 0.0);
+	nd *= nd; // nd^2
+	return nd * pow(max(distance, 0.0001), -decay);
+}
+
+bool compute_light_vector(uint light, vec3 pos, out float attenuation, out vec3 light_pos) {
+	if (lights.data[light].type == LIGHT_TYPE_DIRECTIONAL) {
+		light_pos = pos - lights.data[light].direction * length(vec3(params.limits));
+		attenuation = 1.0;
+
+	} else {
+		light_pos = lights.data[light].position;
+		float distance = length(pos - light_pos);
+		if (distance >= lights.data[light].radius) {
+			return false;
+		}
+
+		attenuation = get_omni_attenuation(
+				distance * params.cell_size,
+				1.0 / (lights.data[light].radius * params.cell_size),
+				lights.data[light].attenuation);
+
+		if (lights.data[light].type == LIGHT_TYPE_SPOT) {
+			vec3 rel = normalize(pos - light_pos);
+			float cos_spot_angle = lights.data[light].cos_spot_angle;
+			float cos_angle = dot(rel, lights.data[light].direction);
+			if (cos_angle < cos_spot_angle) {
+				return false;
+			}
+
+			float scos = max(cos_angle, cos_spot_angle);
+			float spot_rim = max(0.0001, (1.0 - scos) / (1.0 - cos_spot_angle));
+			attenuation *= 1.0 - pow(spot_rim, lights.data[light].inv_spot_attenuation);
+		}
+	}
+
+	return true;
+}
+
+float get_normal_advance(vec3 p_normal) {
+	vec3 normal = p_normal;
+	vec3 unorm = abs(normal);
+
+	if ((unorm.x >= unorm.y) && (unorm.x >= unorm.z)) {
+		// x code
+		unorm = normal.x > 0.0 ? vec3(1.0, 0.0, 0.0) : vec3(-1.0, 0.0, 0.0);
+	} else if ((unorm.y > unorm.x) && (unorm.y >= unorm.z)) {
+		// y code
+		unorm = normal.y > 0.0 ? vec3(0.0, 1.0, 0.0) : vec3(0.0, -1.0, 0.0);
+	} else if ((unorm.z > unorm.x) && (unorm.z > unorm.y)) {
+		// z code
+		unorm = normal.z > 0.0 ? vec3(0.0, 0.0, 1.0) : vec3(0.0, 0.0, -1.0);
+	} else {
+		// oh-no we messed up code
+		// has to be
+		unorm = vec3(1.0, 0.0, 0.0);
+	}
+
+	return 1.0 / dot(normal, unorm);
+}
+
+void clip_segment(vec4 plane, vec3 begin, inout vec3 end) {
+	vec3 segment = begin - end;
+	float den = dot(plane.xyz, segment);
+
+	//printf("den is %i\n",den);
+	if (den < 0.0001) {
+		return;
+	}
+
+	float dist = (dot(plane.xyz, begin) - plane.w) / den;
+
+	if (dist < 0.0001 || dist > 1.0001) {
+		return;
+	}
+
+	end = begin + segment * -dist;
+}
+
+bool compute_light_at_pos(uint index, vec3 pos, vec3 normal, inout vec3 light, inout vec3 light_dir) {
+	float attenuation;
+	vec3 light_pos;
+
+	if (!compute_light_vector(index, pos, attenuation, light_pos)) {
+		return false;
+	}
+
+	light_dir = normalize(pos - light_pos);
+
+	if (attenuation < 0.01 || (length(normal) > 0.2 && dot(normal, light_dir) >= 0)) {
+		return false; //not facing the light, or attenuation is near zero
+	}
+
+	if (lights.data[index].has_shadow) {
+		float distance_adv = get_normal_advance(light_dir);
+
+		vec3 to = pos;
+		if (length(normal) > 0.2) {
+			to += normal * distance_adv * 0.51;
+		} else {
+			to -= sign(light_dir) * 0.45; //go near the edge towards the light direction to avoid self occlusion
+		}
+
+		//clip
+		clip_segment(mix(vec4(-1.0, 0.0, 0.0, 0.0), vec4(1.0, 0.0, 0.0, float(params.limits.x - 1)), bvec4(light_dir.x < 0.0)), to, light_pos);
+		clip_segment(mix(vec4(0.0, -1.0, 0.0, 0.0), vec4(0.0, 1.0, 0.0, float(params.limits.y - 1)), bvec4(light_dir.y < 0.0)), to, light_pos);
+		clip_segment(mix(vec4(0.0, 0.0, -1.0, 0.0), vec4(0.0, 0.0, 1.0, float(params.limits.z - 1)), bvec4(light_dir.z < 0.0)), to, light_pos);
+
+		float distance = length(to - light_pos);
+		if (distance < 0.1) {
+			return false; // hit
+		}
+
+		distance += distance_adv - mod(distance, distance_adv); //make it reach the center of the box always
+		light_pos = to - light_dir * distance;
+
+		//from -= sign(light_dir)*0.45; //go near the edge towards the light direction to avoid self occlusion
+
+		/*float dist = raymarch(distance,distance_adv,light_pos,light_dir);
+
+		if (dist > distance_adv) {
+			return false;
+		}
+
+		attenuation *= 1.0 - smoothstep(0.1*distance_adv,distance_adv,dist);
+		*/
+
+		float occlusion = raymarch(distance, distance_adv, light_pos, light_dir);
+
+		if (occlusion == 0.0) {
+			return false;
+		}
+
+		attenuation *= occlusion; //1.0 - smoothstep(0.1*distance_adv,distance_adv,dist);
+	}
+
+	light = lights.data[index].color * attenuation * lights.data[index].energy;
+	return true;
+}
+
+#endif // MODE COMPUTE LIGHT
+
+void main() {
+#ifndef MODE_DYNAMIC
+
+	uint cell_index = gl_GlobalInvocationID.x;
+	if (cell_index >= params.cell_count) {
+		return;
+	}
+	cell_index += params.cell_offset;
+
+	uvec3 posu = uvec3(cell_data.data[cell_index].position & 0x7FF, (cell_data.data[cell_index].position >> 11) & 0x3FF, cell_data.data[cell_index].position >> 21);
+	vec4 albedo = unpackUnorm4x8(cell_data.data[cell_index].albedo);
+
+#endif
+
+	/////////////////COMPUTE LIGHT///////////////////////////////
+
+#ifdef MODE_COMPUTE_LIGHT
+
+	vec3 pos = vec3(posu) + vec3(0.5);
+
+	vec3 emission = vec3(uvec3(cell_data.data[cell_index].emission & 0x1ff, (cell_data.data[cell_index].emission >> 9) & 0x1ff, (cell_data.data[cell_index].emission >> 18) & 0x1ff)) * pow(2.0, float(cell_data.data[cell_index].emission >> 27) - 15.0 - 9.0);
+	vec3 normal = unpackSnorm4x8(cell_data.data[cell_index].normal).xyz;
+
+	vec3 accum = vec3(0.0);
+
+	for (uint i = 0; i < params.light_count; i++) {
+		vec3 light;
+		vec3 light_dir;
+		if (!compute_light_at_pos(i, pos, normal.xyz, light, light_dir)) {
+			continue;
+		}
+
+		light *= albedo.rgb;
+
+		if (length(normal) > 0.2) {
+			accum += max(0.0, dot(normal, -light_dir)) * light;
+		} else {
+			//all directions
+			accum += light;
+		}
+	}
+
+	outputs.data[cell_index] = vec4(accum + emission, 0.0);
+
+#endif //MODE_COMPUTE_LIGHT
+
+	/////////////////SECOND BOUNCE///////////////////////////////
+
+#ifdef MODE_SECOND_BOUNCE
+	vec3 pos = vec3(posu) + vec3(0.5);
+	ivec3 ipos = ivec3(posu);
+	vec4 normal = unpackSnorm4x8(cell_data.data[cell_index].normal);
+
+	vec3 accum = outputs.data[cell_index].rgb;
+
+	if (length(normal.xyz) > 0.2) {
+		vec3 v0 = abs(normal.z) < 0.999 ? vec3(0.0, 0.0, 1.0) : vec3(0.0, 1.0, 0.0);
+		vec3 tangent = normalize(cross(v0, normal.xyz));
+		vec3 bitangent = normalize(cross(tangent, normal.xyz));
+		mat3 normal_mat = mat3(tangent, bitangent, normal.xyz);
+
+#define MAX_CONE_DIRS 6
+
+		vec3 cone_dirs[MAX_CONE_DIRS] = vec3[](
+				vec3(0.0, 0.0, 1.0),
+				vec3(0.866025, 0.0, 0.5),
+				vec3(0.267617, 0.823639, 0.5),
+				vec3(-0.700629, 0.509037, 0.5),
+				vec3(-0.700629, -0.509037, 0.5),
+				vec3(0.267617, -0.823639, 0.5));
+
+		float cone_weights[MAX_CONE_DIRS] = float[](0.25, 0.15, 0.15, 0.15, 0.15, 0.15);
+		float tan_half_angle = 0.577;
+
+		for (int i = 0; i < MAX_CONE_DIRS; i++) {
+			vec3 direction = normal_mat * cone_dirs[i];
+			vec4 color = vec4(0.0);
+			{
+				float dist = 1.5;
+				float max_distance = length(vec3(params.limits));
+				vec3 cell_size = 1.0 / vec3(params.limits);
+
+				while (dist < max_distance && color.a < 0.95) {
+					float diameter = max(1.0, 2.0 * tan_half_angle * dist);
+					vec3 uvw_pos = (pos + dist * direction) * cell_size;
+					float half_diameter = diameter * 0.5;
+					//check if outside, then break
+					//if ( any(greaterThan(abs(uvw_pos - 0.5),vec3(0.5f + half_diameter * cell_size)) ) ) {
+					//	break;
+					//}
+
+					float log2_diameter = log2(diameter);
+					vec4 scolor = textureLod(sampler3D(color_texture, texture_sampler), uvw_pos, log2_diameter);
+					float a = (1.0 - color.a);
+					color += a * scolor;
+					dist += half_diameter;
+				}
+			}
+			color *= cone_weights[i] * vec4(albedo.rgb, 1.0) * params.dynamic_range; //restore range
+			accum += color.rgb;
+		}
+	}
+
+	outputs.data[cell_index] = vec4(accum, 0.0);
+
+#endif // MODE_SECOND_BOUNCE
+
+	/////////////////UPDATE MIPMAPS///////////////////////////////
+
+#ifdef MODE_UPDATE_MIPMAPS
+
+	{
+		vec3 light_accum = vec3(0.0);
+		float count = 0.0;
+		for (uint i = 0; i < 8; i++) {
+			uint child_index = cell_children.data[cell_index].children[i];
+			if (child_index == NO_CHILDREN) {
+				continue;
+			}
+			light_accum += outputs.data[child_index].rgb;
+
+			count += 1.0;
+		}
+
+		float divisor = mix(8.0, count, params.propagation);
+		outputs.data[cell_index] = vec4(light_accum / divisor, 0.0);
+	}
+#endif
+
+	///////////////////WRITE TEXTURE/////////////////////////////
+
+#ifdef MODE_WRITE_TEXTURE
+	{
+		imageStore(color_tex, ivec3(posu), vec4(outputs.data[cell_index].rgb / params.dynamic_range, albedo.a));
+	}
+#endif
+
+	///////////////////DYNAMIC LIGHTING/////////////////////////////
+
+#ifdef MODE_DYNAMIC
+
+	ivec2 pos_xy = ivec2(gl_GlobalInvocationID.xy);
+	if (any(greaterThanEqual(pos_xy, params.rect_size))) {
+		return; //out of bounds
+	}
+
+	ivec2 uv_xy = pos_xy;
+	if (params.flip_x) {
+		uv_xy.x = params.rect_size.x - pos_xy.x - 1;
+	}
+	if (params.flip_y) {
+		uv_xy.y = params.rect_size.y - pos_xy.y - 1;
+	}
+
+#ifdef MODE_DYNAMIC_LIGHTING
+
+	{
+		float z = params.z_base + imageLoad(depth, uv_xy).x * params.z_sign;
+
+		ivec3 pos = params.x_dir * (params.rect_pos.x + pos_xy.x) + params.y_dir * (params.rect_pos.y + pos_xy.y) + abs(params.z_dir) * int(z);
+
+		vec3 normal = normalize(imageLoad(source_normal, uv_xy).xyz * 2.0 - 1.0);
+		normal = vec3(params.x_dir) * normal.x * mix(1.0, -1.0, params.flip_x) + vec3(params.y_dir) * normal.y * mix(1.0, -1.0, params.flip_y) - vec3(params.z_dir) * normal.z;
+
+		vec4 albedo = imageLoad(source_albedo, uv_xy);
+
+		//determine the position in space
+
+		vec3 accum = vec3(0.0);
+		for (uint i = 0; i < params.light_count; i++) {
+			vec3 light;
+			vec3 light_dir;
+			if (!compute_light_at_pos(i, vec3(pos) * params.pos_multiplier, normal, light, light_dir)) {
+				continue;
+			}
+
+			light *= albedo.rgb;
+
+			accum += max(0.0, dot(normal, -light_dir)) * light;
+		}
+
+		accum += imageLoad(emission, uv_xy).xyz;
+
+		imageStore(emission, uv_xy, vec4(accum, albedo.a));
+		imageStore(depth, uv_xy, vec4(z));
+	}
+
+#endif // MODE DYNAMIC LIGHTING
+
+#ifdef MODE_DYNAMIC_SHRINK
+
+	{
+		vec4 accum = vec4(0.0);
+		float accum_z = 0.0;
+		float count = 0.0;
+
+		for (int i = 0; i < 4; i++) {
+			ivec2 ofs = pos_xy * 2 + ivec2(i & 1, i >> 1) - params.prev_rect_ofs;
+			if (any(lessThan(ofs, ivec2(0))) || any(greaterThanEqual(ofs, params.prev_rect_size))) {
+				continue;
+			}
+			if (params.flip_x) {
+				ofs.x = params.prev_rect_size.x - ofs.x - 1;
+			}
+			if (params.flip_y) {
+				ofs.y = params.prev_rect_size.y - ofs.y - 1;
+			}
+
+			vec4 light = imageLoad(source_light, ofs);
+			if (light.a == 0.0) { //ignore empty
+				continue;
+			}
+			accum += light;
+			float z = imageLoad(source_depth, ofs).x;
+			accum_z += z * 0.5; //shrink half too
+			count += 1.0;
+		}
+
+		if (params.on_mipmap) {
+			accum.rgb /= mix(8.0, count, params.propagation);
+			accum.a /= 8.0;
+		} else {
+			accum /= 4.0;
+		}
+
+		if (count == 0.0) {
+			accum_z = 0.0; //avoid nan
+		} else {
+			accum_z /= count;
+		}
+
+#ifdef MODE_DYNAMIC_SHRINK_WRITE
+
+		imageStore(light, uv_xy, accum);
+		imageStore(depth, uv_xy, vec4(accum_z));
+#endif
+
+#ifdef MODE_DYNAMIC_SHRINK_PLOT
+
+		if (accum.a < 0.001) {
+			return; //do not blit if alpha is too low
+		}
+
+		ivec3 pos = params.x_dir * (params.rect_pos.x + pos_xy.x) + params.y_dir * (params.rect_pos.y + pos_xy.y) + abs(params.z_dir) * int(accum_z);
+
+		float z_frac = fract(accum_z);
+
+		for (int i = 0; i < 2; i++) {
+			ivec3 pos3d = pos + abs(params.z_dir) * i;
+			if (any(lessThan(pos3d, ivec3(0))) || any(greaterThanEqual(pos3d, params.limits))) {
+				//skip if offlimits
+				continue;
+			}
+			vec4 color_blit = accum * (i == 0 ? 1.0 - z_frac : z_frac);
+			vec4 color = imageLoad(color_texture, pos3d);
+			color.rgb *= params.dynamic_range;
+
+#if 0
+			color.rgb = mix(color.rgb,color_blit.rgb,color_blit.a);
+			color.a+=color_blit.a;
+#else
+
+			float sa = 1.0 - color_blit.a;
+			vec4 result;
+			result.a = color.a * sa + color_blit.a;
+			if (result.a == 0.0) {
+				result = vec4(0.0);
+			} else {
+				result.rgb = (color.rgb * color.a * sa + color_blit.rgb * color_blit.a) / result.a;
+				color = result;
+			}
+
+#endif
+			color.rgb /= params.dynamic_range;
+			imageStore(color_texture, pos3d, color);
+			//imageStore(color_texture,pos3d,vec4(1,1,1,1));
+		}
+#endif // MODE_DYNAMIC_SHRINK_PLOT
+	}
+#endif
+
+#endif // MODE DYNAMIC
+}

servers/rendering/renderer_rd/shaders/environment/voxel_gi_debug.glsl

@@ -0,0 +1,184 @@
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+struct CellData {
+	uint position; // xyz 10 bits
+	uint albedo; //rgb albedo
+	uint emission; //rgb normalized with e as multiplier
+	uint normal; //RGB normal encoded
+};
+
+layout(set = 0, binding = 1, std140) buffer CellDataBuffer {
+	CellData data[];
+}
+cell_data;
+
+layout(set = 0, binding = 2) uniform texture3D color_tex;
+
+layout(set = 0, binding = 3) uniform sampler tex_sampler;
+
+layout(push_constant, std430) uniform Params {
+	mat4 projection;
+	uint cell_offset;
+	float dynamic_range;
+	float alpha;
+	uint level;
+	ivec3 bounds;
+	uint pad;
+}
+params;
+
+layout(location = 0) out vec4 color_interp;
+
+void main() {
+	const vec3 cube_triangles[36] = vec3[](
+			vec3(-1.0f, -1.0f, -1.0f),
+			vec3(-1.0f, -1.0f, 1.0f),
+			vec3(-1.0f, 1.0f, 1.0f),
+			vec3(1.0f, 1.0f, -1.0f),
+			vec3(-1.0f, -1.0f, -1.0f),
+			vec3(-1.0f, 1.0f, -1.0f),
+			vec3(1.0f, -1.0f, 1.0f),
+			vec3(-1.0f, -1.0f, -1.0f),
+			vec3(1.0f, -1.0f, -1.0f),
+			vec3(1.0f, 1.0f, -1.0f),
+			vec3(1.0f, -1.0f, -1.0f),
+			vec3(-1.0f, -1.0f, -1.0f),
+			vec3(-1.0f, -1.0f, -1.0f),
+			vec3(-1.0f, 1.0f, 1.0f),
+			vec3(-1.0f, 1.0f, -1.0f),
+			vec3(1.0f, -1.0f, 1.0f),
+			vec3(-1.0f, -1.0f, 1.0f),
+			vec3(-1.0f, -1.0f, -1.0f),
+			vec3(-1.0f, 1.0f, 1.0f),
+			vec3(-1.0f, -1.0f, 1.0f),
+			vec3(1.0f, -1.0f, 1.0f),
+			vec3(1.0f, 1.0f, 1.0f),
+			vec3(1.0f, -1.0f, -1.0f),
+			vec3(1.0f, 1.0f, -1.0f),
+			vec3(1.0f, -1.0f, -1.0f),
+			vec3(1.0f, 1.0f, 1.0f),
+			vec3(1.0f, -1.0f, 1.0f),
+			vec3(1.0f, 1.0f, 1.0f),
+			vec3(1.0f, 1.0f, -1.0f),
+			vec3(-1.0f, 1.0f, -1.0f),
+			vec3(1.0f, 1.0f, 1.0f),
+			vec3(-1.0f, 1.0f, -1.0f),
+			vec3(-1.0f, 1.0f, 1.0f),
+			vec3(1.0f, 1.0f, 1.0f),
+			vec3(-1.0f, 1.0f, 1.0f),
+			vec3(1.0f, -1.0f, 1.0f));
+
+	vec3 vertex = cube_triangles[gl_VertexIndex] * 0.5 + 0.5;
+#ifdef MODE_DEBUG_LIGHT_FULL
+	uvec3 posu = uvec3(gl_InstanceIndex % params.bounds.x, (gl_InstanceIndex / params.bounds.x) % params.bounds.y, gl_InstanceIndex / (params.bounds.y * params.bounds.x));
+#else
+	uint cell_index = gl_InstanceIndex + params.cell_offset;
+
+	uvec3 posu = uvec3(cell_data.data[cell_index].position & 0x7FF, (cell_data.data[cell_index].position >> 11) & 0x3FF, cell_data.data[cell_index].position >> 21);
+#endif
+
+#ifdef MODE_DEBUG_EMISSION
+	color_interp.xyz = vec3(uvec3(cell_data.data[cell_index].emission & 0x1ff, (cell_data.data[cell_index].emission >> 9) & 0x1ff, (cell_data.data[cell_index].emission >> 18) & 0x1ff)) * pow(2.0, float(cell_data.data[cell_index].emission >> 27) - 15.0 - 9.0);
+#endif
+
+#ifdef MODE_DEBUG_COLOR
+	color_interp.xyz = unpackUnorm4x8(cell_data.data[cell_index].albedo).xyz;
+#endif
+
+#ifdef MODE_DEBUG_LIGHT
+	color_interp = texelFetch(sampler3D(color_tex, tex_sampler), ivec3(posu), int(params.level));
+	color_interp.xyz *params.dynamic_range;
+#endif
+
+	float scale = (1 << params.level);
+
+	gl_Position = params.projection * vec4((vec3(posu) + vertex) * scale, 1.0);
+
+#ifdef MODE_DEBUG_LIGHT_FULL
+	if (color_interp.a == 0.0) {
+		gl_Position = vec4(0.0); //force clip and not draw
+	}
+#else
+	color_interp.a = params.alpha;
+#endif
+}
+
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(location = 0) in vec4 color_interp;
+layout(location = 0) out vec4 frag_color;
+
+void main() {
+	frag_color = color_interp;
+
+#ifdef MODE_DEBUG_LIGHT_FULL
+
+	//there really is no alpha, so use dither
+
+	int x = int(gl_FragCoord.x) % 4;
+	int y = int(gl_FragCoord.y) % 4;
+	int index = x + y * 4;
+	float limit = 0.0;
+	if (x < 8) {
+		if (index == 0) {
+			limit = 0.0625;
+		}
+		if (index == 1) {
+			limit = 0.5625;
+		}
+		if (index == 2) {
+			limit = 0.1875;
+		}
+		if (index == 3) {
+			limit = 0.6875;
+		}
+		if (index == 4) {
+			limit = 0.8125;
+		}
+		if (index == 5) {
+			limit = 0.3125;
+		}
+		if (index == 6) {
+			limit = 0.9375;
+		}
+		if (index == 7) {
+			limit = 0.4375;
+		}
+		if (index == 8) {
+			limit = 0.25;
+		}
+		if (index == 9) {
+			limit = 0.75;
+		}
+		if (index == 10) {
+			limit = 0.125;
+		}
+		if (index == 11) {
+			limit = 0.625;
+		}
+		if (index == 12) {
+			limit = 1.0;
+		}
+		if (index == 13) {
+			limit = 0.5;
+		}
+		if (index == 14) {
+			limit = 0.875;
+		}
+		if (index == 15) {
+			limit = 0.375;
+		}
+	}
+	if (frag_color.a < limit) {
+		discard;
+	}
+#endif
+}

servers/rendering/renderer_rd/shaders/environment/voxel_gi_sdf.glsl

@@ -0,0 +1,180 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(local_size_x = 4, local_size_y = 4, local_size_z = 4) in;
+
+#define MAX_DISTANCE 100000.0
+
+#define NO_CHILDREN 0xFFFFFFFF
+
+struct CellChildren {
+	uint children[8];
+};
+
+layout(set = 0, binding = 1, std430) buffer CellChildrenBuffer {
+	CellChildren data[];
+}
+cell_children;
+
+struct CellData {
+	uint position; // xyz 10 bits
+	uint albedo; //rgb albedo
+	uint emission; //rgb normalized with e as multiplier
+	uint normal; //RGB normal encoded
+};
+
+layout(set = 0, binding = 2, std430) buffer CellDataBuffer {
+	CellData data[];
+}
+cell_data;
+
+layout(r8ui, set = 0, binding = 3) uniform restrict writeonly uimage3D sdf_tex;
+
+layout(push_constant, std430) uniform Params {
+	uint offset;
+	uint end;
+	uint pad0;
+	uint pad1;
+}
+params;
+
+void main() {
+	vec3 pos = vec3(gl_GlobalInvocationID);
+	float closest_dist = MAX_DISTANCE;
+
+	for (uint i = params.offset; i < params.end; i++) {
+		vec3 posu = vec3(uvec3(cell_data.data[i].position & 0x7FF, (cell_data.data[i].position >> 11) & 0x3FF, cell_data.data[i].position >> 21));
+		float dist = length(pos - posu);
+		if (dist < closest_dist) {
+			closest_dist = dist;
+		}
+	}
+
+	uint dist_8;
+
+	if (closest_dist < 0.0001) { // same cell
+		dist_8 = 0; //equals to -1
+	} else {
+		dist_8 = clamp(uint(closest_dist), 0, 254) + 1; //conservative, 0 is 1, so <1 is considered solid
+	}
+
+	imageStore(sdf_tex, ivec3(gl_GlobalInvocationID), uvec4(dist_8));
+	//imageStore(sdf_tex,pos,uvec4(pos*2,0));
+}
+
+#if 0
+layout(push_constant, std430) uniform Params {
+	ivec3 limits;
+	uint stack_size;
+}
+params;
+
+float distance_to_aabb(ivec3 pos, ivec3 aabb_pos, ivec3 aabb_size) {
+	vec3 delta = vec3(max(ivec3(0), max(aabb_pos - pos, pos - (aabb_pos + aabb_size - ivec3(1)))));
+	return length(delta);
+}
+
+void main() {
+	ivec3 pos = ivec3(gl_GlobalInvocationID);
+
+	uint stack[10] = uint[](0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
+	uint stack_indices[10] = uint[](0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
+	ivec3 stack_positions[10] = ivec3[](ivec3(0), ivec3(0), ivec3(0), ivec3(0), ivec3(0), ivec3(0), ivec3(0), ivec3(0), ivec3(0), ivec3(0));
+
+	const uint cell_orders[8] = uint[](
+			0x11f58d1,
+			0xe2e70a,
+			0xd47463,
+			0xbb829c,
+			0x8d11f5,
+			0x70ae2e,
+			0x463d47,
+			0x29cbb8);
+
+	bool cell_found = false;
+	bool cell_found_exact = false;
+	ivec3 closest_cell_pos;
+	float closest_distance = MAX_DISTANCE;
+	int stack_pos = 0;
+
+	while (true) {
+		uint index = stack_indices[stack_pos] >> 24;
+
+		if (index == 8) {
+			//go up
+			if (stack_pos == 0) {
+				break; //done going through octree
+			}
+			stack_pos--;
+			continue;
+		}
+
+		stack_indices[stack_pos] = (stack_indices[stack_pos] & ((1 << 24) - 1)) | ((index + 1) << 24);
+
+		uint cell_index = (stack_indices[stack_pos] >> (index * 3)) & 0x7;
+		uint child_cell = cell_children.data[stack[stack_pos]].children[cell_index];
+
+		if (child_cell == NO_CHILDREN) {
+			continue;
+		}
+
+		ivec3 child_cell_size = params.limits >> (stack_pos + 1);
+		ivec3 child_cell_pos = stack_positions[stack_pos];
+
+		child_cell_pos += mix(ivec3(0), child_cell_size, bvec3(uvec3(index & 1, index & 2, index & 4) != uvec3(0)));
+
+		bool is_leaf = stack_pos == (params.stack_size - 2);
+
+		if (child_cell_pos == pos && is_leaf) {
+			//we may actually end up in the exact cell.
+			//if this happens, just abort
+			cell_found_exact = true;
+			break;
+		}
+
+		if (cell_found) {
+			//discard by distance
+			float distance = distance_to_aabb(pos, child_cell_pos, child_cell_size);
+			if (distance >= closest_distance) {
+				continue; //pointless, just test next child
+			} else if (is_leaf) {
+				//closer than what we have AND end of stack, save and continue
+				closest_cell_pos = child_cell_pos;
+				closest_distance = distance;
+				continue;
+			}
+		} else if (is_leaf) {
+			//first solid cell we find, save and continue
+			closest_distance = distance_to_aabb(pos, child_cell_pos, child_cell_size);
+			closest_cell_pos = child_cell_pos;
+			cell_found = true;
+			continue;
+		}
+
+		bvec3 direction = greaterThan((pos - (child_cell_pos + (child_cell_size >> 1))), ivec3(0));
+		uint cell_order = 0;
+		cell_order |= mix(0, 1, direction.x);
+		cell_order |= mix(0, 2, direction.y);
+		cell_order |= mix(0, 4, direction.z);
+
+		stack[stack_pos + 1] = child_cell;
+		stack_indices[stack_pos + 1] = cell_orders[cell_order]; //start counting
+		stack_positions[stack_pos + 1] = child_cell_pos;
+		stack_pos++; //go up stack
+	}
+
+	uint dist_8;
+
+	if (cell_found_exact) {
+		dist_8 = 0; //equals to -1
+	} else {
+		float closest_distance = length(vec3(pos - closest_cell_pos));
+		dist_8 = clamp(uint(closest_distance), 0, 254) + 1; //conservative, 0 is 1, so <1 is considered solid
+	}
+
+	imageStore(sdf_tex, pos, uvec4(dist_8));
+}
+#endif